diff --git a/include/wx/convauto.h b/include/wx/convauto.h
index 3da6c6adc4..d7ed45592f 100644
--- a/include/wx/convauto.h
+++ b/include/wx/convauto.h
@@ -78,6 +78,8 @@ public:
 
     virtual size_t GetMBNulLen() const wxOVERRIDE { return m_conv->GetMBNulLen(); }
 
+    virtual bool IsUTF8() const wxOVERRIDE { return m_conv && m_conv->IsUTF8(); }
+
     virtual wxMBConv *Clone() const wxOVERRIDE { return new wxConvAuto(*this); }
 
     // return the BOM type of this buffer
@@ -91,6 +93,14 @@ public:
         return m_bomType;
     }
 
+    wxFontEncoding GetEncoding() const;
+
+    // Return true if the fall-back encoding is used
+    bool IsUsingFallbackEncoding() const
+    {
+        return m_ownsConv && m_bomType == wxBOM_None;
+    }
+
 private:
     // common part of all ctors
     void Init()
diff --git a/include/wx/private/unicode.h b/include/wx/private/unicode.h
new file mode 100644
index 0000000000..6c81c23504
--- /dev/null
+++ b/include/wx/private/unicode.h
@@ -0,0 +1,16 @@
+/////////////////////////////////////////////////////////////////////////////
+// Name:        wx/private/unicode.h
+// Purpose:     Unicode private declsrations
+// Author:      Pavel Tyunin
+// Created:     2020-10-06
+// Copyright:   (c) 2020 Pavel Tyunin
+// Licence:     wxWindows licence
+/////////////////////////////////////////////////////////////////////////////
+
+#ifndef _WX_PRIVATE_UNICODEH__
+#define _WX_PRIVATE_UNICODEH__
+
+// this table gives the length of the UTF-8 encoding from its first character:
+extern const unsigned char tableUtf8Lengths[256];
+
+#endif // _WX_PRIVATE_UNICODEH__
diff --git a/include/wx/stringops.h b/include/wx/stringops.h
index 150554f341..dd46e6616c 100644
--- a/include/wx/stringops.h
+++ b/include/wx/stringops.h
@@ -94,15 +94,15 @@ struct WXDLLIMPEXP_BASE wxStringOperationsUtf8
         return (c <= 0x7F) || (c >= 0xC2 && c <= 0xF4);
     }
 
-    // table of offsets to skip forward when iterating over UTF-8 sequence
-    static const unsigned char ms_utf8IterTable[256];
+    // returns offset to skip forward when iterating over UTF-8 sequence
+    static unsigned char GetUTF8IterOffset(unsigned char c);
 
 
     template<typename Iterator>
     static void IncIter(Iterator& i)
     {
         wxASSERT( IsValidUtf8LeadByte(*i) );
-        i += ms_utf8IterTable[(unsigned char)*i];
+        i += GetUTF8IterOffset(*i);
     }
 
     template<typename Iterator>
@@ -178,7 +178,7 @@ struct WXDLLIMPEXP_BASE wxStringOperationsUtf8
     static size_t GetUtf8CharLength(char c)
     {
         wxASSERT( IsValidUtf8LeadByte(c) );
-        return ms_utf8IterTable[(unsigned char)c];
+        return GetUTF8IterOffset(c);
     }
 
     // decodes single UTF-8 character from UTF-8 string
diff --git a/interface/wx/convauto.h b/interface/wx/convauto.h
index 7ddfb26927..324b5b24e7 100644
--- a/interface/wx/convauto.h
+++ b/interface/wx/convauto.h
@@ -146,6 +146,22 @@ public:
     */
     wxBOM GetBOM() const;
 
+    /**
+        Return the detected encoding
+
+        Returns @c wxFONTENCODING_MAX if called before the first use.
+
+        @since 3.1.5
+    */
+    wxBOM GetEncoding() const;
+
+    /**
+        Check if the fall-back encoding is used.
+
+        @since 3.1.5
+    */
+    bool IsUsingFallbackEncoding() const;
+
     /**
         Return a pointer to the characters that makes up this BOM.
 
diff --git a/src/common/convauto.cpp b/src/common/convauto.cpp
index 952b4455f5..708096de5d 100644
--- a/src/common/convauto.cpp
+++ b/src/common/convauto.cpp
@@ -23,6 +23,7 @@
 #endif
 
 #include "wx/convauto.h"
+#include "wx/private/unicode.h"
 
 // we use latin1 by default as it seems the least bad choice: the files we need
 // to detect input of don't always come from the user system (they are often
@@ -266,6 +267,23 @@ bool wxConvAuto::InitFromInput(const char *src, size_t len)
     return true;
 }
 
+// checks if the input can be the beginning of a valid UTF-8 sequence
+static bool wxCanBeUTF8SequencePrefix(const char *src, size_t len)
+{
+    size_t i = 0;
+    unsigned char l = tableUtf8Lengths[(unsigned char)src[i]];
+    if ( !l )
+        return false; // invalid leading byte
+    while ( --l )
+    {
+        if ( ++i == len )
+            return true; // truncated sequence
+        if ( (src[i] & 0xC0) != 0x80 )
+            return false; // invalid continuation byte
+    }
+    return false; // complete sequence
+}
+
 size_t
 wxConvAuto::ToWChar(wchar_t *dst, size_t dstLen,
                     const char *src, size_t srcLen) const
@@ -307,25 +325,28 @@ wxConvAuto::ToWChar(wchar_t *dst, size_t dstLen,
 
     // try to convert using the auto-detected encoding
     size_t rc = m_conv->ToWChar(dst, dstLen, src, srcLen);
-    if ( rc == wxCONV_FAILED && m_bomType == wxBOM_None )
+    if ( rc == wxCONV_FAILED && m_bomType == wxBOM_None && !m_ownsConv )
     {
         // we may need more bytes before we can decode the input, don't switch
         // to the fall-back conversion in this case as it would prevent us from
         // decoding UTF-8 input when fed it byte by byte, as done by
         // wxTextInputStream, for example
-        if ( srcLen < m_conv->GetMaxCharLen() )
+        // up to 2 extra bytes are needed for inputs that start with null bytes
+        // that look like the start of UTF-32BE BOM, but can be in UTF-8 too
+        size_t nNull = 0;
+        if ( srcLen != wxNO_LEN && srcLen >= 2 && !src[0] )
+            nNull = ( src[1]? 1 : 2 );
+        if ( srcLen < nNull + m_conv->GetMaxCharLen() &&
+             wxCanBeUTF8SequencePrefix(src + nNull, srcLen - nNull) )
             return wxCONV_FAILED;
 
         // if the conversion failed but we didn't really detect anything and
         // simply tried UTF-8 by default, retry it using the fall-back
+        if ( m_encDefault == wxFONTENCODING_DEFAULT )
+            self->m_encDefault = GetFallbackEncoding();
         if ( m_encDefault != wxFONTENCODING_MAX )
         {
-            if ( m_ownsConv )
-                delete m_conv;
-
-            self->m_conv = new wxCSConv(m_encDefault == wxFONTENCODING_DEFAULT
-                                            ? GetFallbackEncoding()
-                                            : m_encDefault);
+            self->m_conv = new wxCSConv(m_encDefault);
             self->m_ownsConv = true;
 
             rc = m_conv->ToWChar(dst, dstLen, src, srcLen);
@@ -351,3 +372,32 @@ wxConvAuto::FromWChar(char *dst, size_t dstLen,
 
     return m_conv->FromWChar(dst, dstLen, src, srcLen);
 }
+
+wxFontEncoding wxConvAuto::GetEncoding() const
+{
+    switch ( m_bomType )
+    {
+        case wxBOM_UTF32BE:
+            return wxFONTENCODING_UTF32BE;
+        case wxBOM_UTF32LE:
+            return wxFONTENCODING_UTF32LE;
+        case wxBOM_UTF16BE:
+            return wxFONTENCODING_UTF16BE;
+        case wxBOM_UTF16LE:
+            return wxFONTENCODING_UTF16LE;
+        case wxBOM_UTF8:
+            return wxFONTENCODING_UTF8;
+
+        case wxBOM_Unknown:
+        case wxBOM_None:
+            if ( !m_conv )
+                return wxFONTENCODING_MAX;
+            else if ( !m_ownsConv )
+                return wxFONTENCODING_UTF8;
+            else
+                return m_encDefault;
+    }
+
+    wxFAIL_MSG( "unknown BOM type" );
+    return wxFONTENCODING_MAX;
+}
diff --git a/src/common/strconv.cpp b/src/common/strconv.cpp
index ba25dae157..23c2b0a545 100644
--- a/src/common/strconv.cpp
+++ b/src/common/strconv.cpp
@@ -46,6 +46,7 @@
 
 #include "wx/encconv.h"
 #include "wx/fontmap.h"
+#include "wx/private/unicode.h"
 
 #ifdef __DARWIN__
 #include "wx/osx/core/private/strconv_cf.h"
@@ -921,7 +922,7 @@ const wxUint32 wxUnicodePUA = 0x100000;
 const wxUint32 wxUnicodePUAEnd = wxUnicodePUA + 256;
 
 // this table gives the length of the UTF-8 encoding from its first character:
-const unsigned char tableUtf8Lengths[256] = {
+extern const unsigned char tableUtf8Lengths[256] = {
     // single-byte sequences (ASCII):
     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 00..0F
     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 10..1F
diff --git a/src/common/stringops.cpp b/src/common/stringops.cpp
index 85629406a3..84017ae523 100644
--- a/src/common/stringops.cpp
+++ b/src/common/stringops.cpp
@@ -23,6 +23,8 @@
     #include "wx/stringops.h"
 #endif
 
+#include "wx/private/unicode.h"
+
 // ===========================================================================
 // implementation
 // ===========================================================================
@@ -97,40 +99,13 @@ wxWxCharBuffer wxStringOperationsWchar::EncodeNChars(size_t n, const wxUniChar&
 // UTF-8 sequences lengths
 // ---------------------------------------------------------------------------
 
-const unsigned char wxStringOperationsUtf8::ms_utf8IterTable[256] = {
-    // single-byte sequences (ASCII):
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 00..0F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 10..1F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 20..2F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 30..3F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 40..4F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 50..5F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 60..6F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 70..7F
-
-    // these are invalid, we use step 1 to skip
-    // over them (should never happen):
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 80..8F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 90..9F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // A0..AF
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // B0..BF
-    1, 1,                                            // C0,C1
-
-    // two-byte sequences:
-          2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  // C2..CF
-    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  // D0..DF
-
-    // three-byte sequences:
-    3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,  // E0..EF
-
-    // four-byte sequences:
-    4, 4, 4, 4, 4,                                   // F0..F4
-
-    // these are invalid again (5- or 6-byte
-    // sequences and sequences for code points
-    // above U+10FFFF, as restricted by RFC 3629):
-                   1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1   // F5..FF
-};
+unsigned char wxStringOperationsUtf8::GetUTF8IterOffset(unsigned char c)
+{
+    unsigned char l = tableUtf8Lengths[c];
+    if ( !l ) //skip over invalid characters
+        l = 1;
+    return l;
+}
 
 // ---------------------------------------------------------------------------
 // UTF-8 operations
@@ -166,7 +141,7 @@ bool wxStringOperationsUtf8::IsValidUtf8String(const char *str, size_t len)
         {
             // if the string is not NULL-terminated, verify we have enough
             // bytes in it left for current character's encoding:
-            if ( c + ms_utf8IterTable[*c] > end )
+            if ( c + GetUTF8IterOffset(*c) > end )
                 return false;
         }
 
@@ -364,7 +339,7 @@ wxCharBuffer wxStringOperationsUtf8::EncodeNChars(size_t n, const wxUniChar& ch)
 {
     Utf8CharBuffer once(EncodeChar(ch));
     // the IncIter() table can be used to determine the length of ch's encoding:
-    size_t len = ms_utf8IterTable[(unsigned char)once.data[0]];
+    size_t len = GetUTF8IterOffset(once.data[0]);
 
     wxCharBuffer buf(n * len);
     char *ptr = buf.data();
diff --git a/src/common/txtstrm.cpp b/src/common/txtstrm.cpp
index c38f7c29ab..1332bb3cf0 100644
--- a/src/common/txtstrm.cpp
+++ b/src/common/txtstrm.cpp
@@ -97,10 +97,11 @@ wxChar wxTextInputStream::GetChar()
         m_validEnd = 0;
     }
 
-    // We may need to decode up to 4 characters if we have input starting with
-    // 3 BOM-like bytes, but not actually containing a BOM, as decoding it will
-    // only succeed when 4 bytes are read -- and will yield 4 wide characters.
-    wxChar wbuf[4];
+    // We may need to decode up to 6 characters if we have input starting with
+    // 2 null bytes (like in UTF-32BE BOM), and then 3 bytes that look like
+    // the start of UTF-8 sequence, as decoding it will only succeed when
+    // 6 bytes are read -- and will yield 6 wide characters.
+    wxChar wbuf[6];
     for(size_t inlen = 0; inlen < sizeof(m_lastBytes); inlen++)
     {
         if ( inlen >= m_validEnd )
@@ -134,12 +135,13 @@ wxChar wxTextInputStream::GetChar()
                 // one extra byte, the only explanation is that we were using a
                 // wxConvAuto conversion recognizing the initial BOM and that
                 // it couldn't detect the presence or absence of BOM so far,
-                // but now finally has enough data to see that there is none.
-                // As we must have fallen back to Latin-1 in this case, return
-                // just the first byte and keep the other ones for the next
-                // time.
-                m_validBegin = 1;
-                return wbuf[0];
+                // but now finally has enough data to see that there is none, or
+                // it was trying to decode the data as UTF-8 sequence, but now
+                // recognized that it's not valid UTF-8 and switched to fallback.
+                // We don't know how long is the first character or if it's decoded
+                // as 1 or 2 wchar_t characters, so we need to start with 1 byte again.
+                inlen = -1;
+                break;
 
 #if SIZEOF_WCHAR_T == 2
             case 2:
diff --git a/src/common/ustring.cpp b/src/common/ustring.cpp
index 6e1768064b..531ee41b9c 100644
--- a/src/common/ustring.cpp
+++ b/src/common/ustring.cpp
@@ -15,6 +15,7 @@
 #endif
 
 #include "wx/ustring.h"
+#include "wx/private/unicode.h"
 
 #ifndef WX_PRECOMP
     #include "wx/crt.h"
@@ -67,41 +68,6 @@ wxUString &wxUString::assignFromAscii( const char *str, size_type n )
 // UTF-8
 // ----------------------------------------------------------------------------
 
-// this table gives the length of the UTF-8 encoding from its first character:
-const unsigned char tableUtf8Lengths[256] = {
-    // single-byte sequences (ASCII):
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 00..0F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 10..1F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 20..2F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 30..3F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 40..4F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 50..5F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 60..6F
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  // 70..7F
-
-    // these are invalid:
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  // 80..8F
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  // 90..9F
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  // A0..AF
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  // B0..BF
-    0, 0,                                            // C0,C1
-
-    // two-byte sequences:
-          2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  // C2..CF
-    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,  // D0..DF
-
-    // three-byte sequences:
-    3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,  // E0..EF
-
-    // four-byte sequences:
-    4, 4, 4, 4, 4,                                   // F0..F4
-
-    // these are invalid again (5- or 6-byte
-    // sequences and sequences for code points
-    // above U+10FFFF, as restricted by RFC 3629):
-                   0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0   // F5..FF
-};
-
 wxUString &wxUString::assignFromUTF8( const char *str )
 {
     if (!str)
diff --git a/tests/mbconv/convautotest.cpp b/tests/mbconv/convautotest.cpp
index 6a5d5791d1..12e19c21ed 100644
--- a/tests/mbconv/convautotest.cpp
+++ b/tests/mbconv/convautotest.cpp
@@ -34,7 +34,9 @@ public:
 
 private:
     CPPUNIT_TEST_SUITE( ConvAutoTestCase );
+        CPPUNIT_TEST( Init );
         CPPUNIT_TEST( Empty );
+        CPPUNIT_TEST( Encode );
         CPPUNIT_TEST( Short );
         CPPUNIT_TEST( None );
         CPPUNIT_TEST( UTF32LE );
@@ -42,22 +44,53 @@ private:
         CPPUNIT_TEST( UTF16LE );
         CPPUNIT_TEST( UTF16BE );
         CPPUNIT_TEST( UTF8 );
+        CPPUNIT_TEST( UTF8NoBom );
+        CPPUNIT_TEST( Fallback );
+        CPPUNIT_TEST( FallbackMultibyte );
+        CPPUNIT_TEST( FallbackShort );
         CPPUNIT_TEST( StreamUTF8NoBOM );
         CPPUNIT_TEST( StreamUTF8 );
         CPPUNIT_TEST( StreamUTF16LE );
         CPPUNIT_TEST( StreamUTF16BE );
         CPPUNIT_TEST( StreamUTF32LE );
         CPPUNIT_TEST( StreamUTF32BE );
+        CPPUNIT_TEST( StreamFallback );
+        CPPUNIT_TEST( StreamFallbackMultibyte );
     CPPUNIT_TEST_SUITE_END();
 
+    // expected converter state, UTF-8 without BOM by default
+    struct ConvState
+    {
+        ConvState( wxBOM bom = wxBOM_None,
+                   wxFontEncoding enc = wxFONTENCODING_UTF8,
+                   bool fallback = false )
+            : m_bom(bom), m_enc(enc), m_fallback(fallback) {}
+
+        void Check(const wxConvAuto& conv) const
+        {
+            CPPUNIT_ASSERT( conv.GetBOM() == m_bom );
+            CPPUNIT_ASSERT( conv.GetEncoding() == m_enc );
+            CPPUNIT_ASSERT( conv.IsUsingFallbackEncoding() == m_fallback );
+            CPPUNIT_ASSERT( conv.IsUTF8() == (m_enc == wxFONTENCODING_UTF8) );
+        }
+
+        wxBOM m_bom;
+        wxFontEncoding m_enc;
+        bool m_fallback;
+    };
+
     // real test function: check that converting the src multibyte string to
     // wide char using wxConvAuto yields wch as the first result
     //
     // the length of the string may need to be passed explicitly if it has
     // embedded NULs, otherwise it's not necessary
-    void TestFirstChar(const char *src, wchar_t wch, size_t len = wxNO_LEN);
+    void TestFirstChar(const char *src, wchar_t wch, size_t len = wxNO_LEN,
+                       ConvState st = ConvState(),
+                       wxFontEncoding fe = wxFONTENCODING_DEFAULT);
 
+    void Init();
     void Empty();
+    void Encode();
     void Short();
     void None();
     void UTF32LE();
@@ -65,12 +98,17 @@ private:
     void UTF16LE();
     void UTF16BE();
     void UTF8();
+    void UTF8NoBom();
+    void Fallback();
+    void FallbackMultibyte();
+    void FallbackShort();
 
     // test whether two lines of text are converted properly from a stream
     void TestTextStream(const char *src,
                         size_t srclength,
                         const wxString& line1,
-                        const wxString& line2);
+                        const wxString& line2,
+                        wxFontEncoding fe = wxFONTENCODING_DEFAULT);
 
     void StreamUTF8NoBOM();
     void StreamUTF8();
@@ -78,6 +116,8 @@ private:
     void StreamUTF16BE();
     void StreamUTF32LE();
     void StreamUTF32BE();
+    void StreamFallback();
+    void StreamFallbackMultibyte();
 };
 
 // register in the unnamed registry so that these tests are run by default
@@ -90,16 +130,36 @@ CPPUNIT_TEST_SUITE_NAMED_REGISTRATION(ConvAutoTestCase, "ConvAutoTestCase");
 // tests
 // ----------------------------------------------------------------------------
 
-void ConvAutoTestCase::TestFirstChar(const char *src, wchar_t wch, size_t len)
+void ConvAutoTestCase::TestFirstChar(const char *src, wchar_t wch, size_t len,
+                                     ConvState st, wxFontEncoding fe)
 {
-    wxWCharBuffer wbuf = wxConvAuto().cMB2WC(src, len, NULL);
+    wxConvAuto conv(fe);
+    wxWCharBuffer wbuf = conv.cMB2WC(src, len, NULL);
     CPPUNIT_ASSERT( wbuf );
     CPPUNIT_ASSERT_EQUAL( wch, *wbuf );
+    st.Check(conv);
+}
+
+void ConvAutoTestCase::Init()
+{
+    ConvState(wxBOM_Unknown, wxFONTENCODING_MAX).Check(wxConvAuto());
 }
 
 void ConvAutoTestCase::Empty()
 {
-    CPPUNIT_ASSERT( !wxConvAuto().cMB2WC("") );
+    wxConvAuto conv;
+    CPPUNIT_ASSERT( !conv.cMB2WC("") );
+    ConvState(wxBOM_Unknown, wxFONTENCODING_MAX).Check(conv);
+}
+
+void ConvAutoTestCase::Encode()
+{
+    wxConvAuto conv;
+    wxString str = wxString::FromUTF8("\xd0\x9f\xe3\x81\x82");
+    wxCharBuffer buf = conv.cWC2MB(str.wc_str());
+    CPPUNIT_ASSERT( buf );
+    CPPUNIT_ASSERT_EQUAL( str, wxString::FromUTF8(buf) );
+    ConvState(wxBOM_Unknown, wxFONTENCODING_UTF8).Check(conv);
 }
 
 void ConvAutoTestCase::Short()
@@ -114,38 +174,71 @@ void ConvAutoTestCase::None()
 
 void ConvAutoTestCase::UTF32LE()
 {
-    TestFirstChar("\xff\xfe\0\0A\0\0\0", wxT('A'), 8);
+    TestFirstChar("\xff\xfe\0\0A\0\0\0", wxT('A'), 8, ConvState(wxBOM_UTF32LE, wxFONTENCODING_UTF32LE));
 }
 
 void ConvAutoTestCase::UTF32BE()
 {
-    TestFirstChar("\0\0\xfe\xff\0\0\0B", wxT('B'), 8);
+    TestFirstChar("\0\0\xfe\xff\0\0\0B", wxT('B'), 8, ConvState(wxBOM_UTF32BE, wxFONTENCODING_UTF32BE));
 }
 
 void ConvAutoTestCase::UTF16LE()
 {
-    TestFirstChar("\xff\xfeZ\0", wxT('Z'), 4);
+    TestFirstChar("\xff\xfeZ\0", wxT('Z'), 4, ConvState(wxBOM_UTF16LE, wxFONTENCODING_UTF16LE));
 }
 
 void ConvAutoTestCase::UTF16BE()
 {
-    TestFirstChar("\xfe\xff\0Y", wxT('Y'), 4);
+    TestFirstChar("\xfe\xff\0Y", wxT('Y'), 4, ConvState(wxBOM_UTF16BE, wxFONTENCODING_UTF16BE));
 }
 
 void ConvAutoTestCase::UTF8()
 {
 #ifdef wxHAVE_U_ESCAPE
-    TestFirstChar("\xef\xbb\xbf\xd0\x9f", L'\u041f');
+    TestFirstChar("\xef\xbb\xbf\xd0\x9f", L'\u041f', wxNO_LEN, ConvState(wxBOM_UTF8, wxFONTENCODING_UTF8));
 #endif
 }
 
+void ConvAutoTestCase::UTF8NoBom()
+{
+#ifdef wxHAVE_U_ESCAPE
+    TestFirstChar("\xd0\x9f\xe3\x81\x82", L'\u041f', wxNO_LEN, ConvState(wxBOM_None, wxFONTENCODING_UTF8));
+#endif
+}
+
+void ConvAutoTestCase::Fallback()
+{
+#ifdef wxHAVE_U_ESCAPE
+    TestFirstChar("\xbf", L'\u041f', wxNO_LEN,
+                  ConvState(wxBOM_None, wxFONTENCODING_ISO8859_5, true),
+                  wxFONTENCODING_ISO8859_5);
+#endif
+}
+
+void ConvAutoTestCase::FallbackMultibyte()
+{
+#ifdef wxHAVE_U_ESCAPE
+    TestFirstChar("\x84\x50", L'\u041f', wxNO_LEN,
+                  ConvState(wxBOM_None, wxFONTENCODING_CP932, true),
+                  wxFONTENCODING_CP932);
+#endif
+}
+
+void ConvAutoTestCase::FallbackShort()
+{
+    TestFirstChar("\x61\xc4", 'a', 2,
+                  ConvState(wxBOM_None, wxFONTENCODING_ISO8859_5, true),
+                  wxFONTENCODING_ISO8859_5);
+}
+
 void ConvAutoTestCase::TestTextStream(const char *src,
                                       size_t srclength,
                                       const wxString& line1,
-                                      const wxString& line2)
+                                      const wxString& line2,
+                                      wxFontEncoding fe)
 {
     wxMemoryInputStream instream(src, srclength);
-    wxTextInputStream text(instream);
+    wxTextInputStream text(instream, wxT(" \t"), wxConvAuto(fe));
 
     CPPUNIT_ASSERT_EQUAL( line1, text.ReadLine() );
     CPPUNIT_ASSERT_EQUAL( line2, text.ReadLine() );
@@ -166,16 +259,8 @@ const wxString line2 = wxString::FromUTF8("\xce\xb2");
 
 void ConvAutoTestCase::StreamUTF8NoBOM()
 {
-    // currently this test doesn't work because without the BOM wxConvAuto
-    // decides that the string is in Latin-1 after finding the first (but not
-    // the two subsequent ones which are part of the same UTF-8 sequence!)
-    // 8-bit character
-    //
-    // FIXME: we need to fix this at wxTextInputStream level, see #11570
-#if 0
     TestTextStream("\x61\xE3\x81\x82\x0A\xCE\xB2",
                    7, line1, line2);
-#endif
 }
 
 void ConvAutoTestCase::StreamUTF8()
@@ -210,4 +295,17 @@ void ConvAutoTestCase::StreamUTF32BE()
                    20, line1, line2);
 }
 
+void ConvAutoTestCase::StreamFallback()
+{
+    TestTextStream("\x61\xbf\x0A\xe0",
+                   4, wxString::FromUTF8("a\xd0\x9f"), wxString::FromUTF8("\xd1\x80"),
+                   wxFONTENCODING_ISO8859_5);
+}
+
+void ConvAutoTestCase::StreamFallbackMultibyte()
+{
+    TestTextStream("\x61\x82\xa0\x0A\x83\xc0",
+                   6, line1, line2, wxFONTENCODING_CP932);
+}
+
 #endif // wxUSE_UNICODE
diff --git a/tests/streams/textstreamtest.cpp b/tests/streams/textstreamtest.cpp
index edb6eaa8a2..c6497b23f2 100644
--- a/tests/streams/textstreamtest.cpp
+++ b/tests/streams/textstreamtest.cpp
@@ -324,6 +324,46 @@ TEST_CASE("wxTextInputStream::GetChar", "[text][input][stream][char]")
         REQUIRE( tis.GetChar() == 0x00 );
         CHECK( tis.GetInputStream().Eof() );
     }
+
+    // Two null bytes that look like the start of UTF-32BE BOM,
+    // followed by 4 byte UTF-8 sequence.
+    // Needs wxConvAuto to not switch to fallback on <6 bytes.
+    SECTION("UTF8-with-nulls")
+    {
+        const wxUint8 buf[] = { 0x00, 0x00, 0xf0, 0x90, 0x8c, 0x98 };
+        wxMemoryInputStream mis(buf, sizeof(buf));
+        wxTextInputStream tis(mis);
+
+        wxCharTypeBuffer<wxChar> e = wxString::FromUTF8((char*)buf, sizeof(buf))
+                                     .tchar_str<wxChar>();
+        for ( size_t i = 0; i < e.length(); ++i )
+        {
+            INFO("i = " << i);
+            REQUIRE( tis.GetChar() == e[i] );
+        }
+        REQUIRE( tis.GetChar() == 0x00 );
+        CHECK( tis.GetInputStream().Eof() );
+    }
+
+    // Two null bytes that look like the start of UTF-32BE BOM,
+    // then 3 bytes that look like the start of UTF-8 sequence.
+    // Needs 6 character output buffer in GetChar().
+    SECTION("almost-UTF8-with-nulls")
+    {
+        const wxUint8 buf[] = { 0x00, 0x00, 0xf0, 0x90, 0x8c, 0xe0 };
+        wxMemoryInputStream mis(buf, sizeof(buf));
+        wxTextInputStream tis(mis);
+
+        wxCharTypeBuffer<wxChar> e = wxString((char*)buf, wxCSConv(wxFONTENCODING_ISO8859_1),
+                                              sizeof(buf)).tchar_str<wxChar>();
+        for ( size_t i = 0; i < e.length(); ++i )
+        {
+            INFO("i = " << i);
+            REQUIRE( tis.GetChar() == e[i] );
+        }
+        REQUIRE( tis.GetChar() == 0x00 );
+        CHECK( tis.GetInputStream().Eof() );
+    }
 }
 
 #endif // wxUSE_UNICODE