diff --git a/src/kits/textencoding/utf8_conversions.cpp b/src/kits/textencoding/utf8_conversions.cpp
index e322828695..86ee58af73 100644
--- a/src/kits/textencoding/utf8_conversions.cpp
+++ b/src/kits/textencoding/utf8_conversions.cpp
@@ -4,17 +4,21 @@
 #include <CharacterSetRoster.h>
 #include <Errors.h>
 #include <errno.h>
+#include <stdio.h>
 
 using namespace BPrivate;
 
 typedef char ** input_buffer_t;
 
+int iconvctl (iconv_t icd, int request, void* argument);
+
 status_t
 convert_encoding(const char * from, const char * to,
                  const char * src, int32 * srcLen,
                  char * dst, int32 * dstLen,
-                 int32 * state)
+                 int32 * state, char substitute)
 {
+	status_t status;
 	if (*srcLen == 0) {
 		// nothing to do!
 		return B_OK;
@@ -32,23 +36,55 @@ convert_encoding(const char * from, const char * to,
 	input_buffer_t inputBuffer = const_cast<input_buffer_t>(&src);
 	size_t inputLeft = *srcLen;
 	size_t outputLeft = *dstLen;
-	size_t nonReversibleConversions = iconv(conversion,inputBuffer,&inputLeft,&dst,&outputLeft);
+	do {
+		size_t nonReversibleConversions = iconv(conversion,inputBuffer,&inputLeft,&dst,&outputLeft);
+		if (nonReversibleConversions == (size_t)-1) {
+			switch (errno) {
+			case EILSEQ: // unable to generate a corresponding character
+				{
+				// discard the input character
+				const int one = 1, zero = 0;
+				iconvctl(conversion,ICONV_SET_DISCARD_ILSEQ,(void*)&one);
+				iconv(conversion,inputBuffer,&inputLeft,&dst,&outputLeft);
+				iconvctl(conversion,ICONV_SET_DISCARD_ILSEQ,(void*)&zero);
+				// prepare to convert the substitute character to target encoding
+				char * original = new char[1];
+				original[0] = substitute;
+				size_t len = 1;
+				char * copy = original;
+				// Perform the conversion
+				// We ignore any errors during this as part of robustness/best-effort
+				// We use ISO-8859-1 as a source because it is a single byte encoding
+				// It also overlaps UTF-8 for the lower 128 characters.  It is also
+				// likely to have a mapping to almost any target encoding.
+				iconv_t iso8859_1to = iconv_open(to,"ISO-8859-1");
+				if (iso8859_1to != (iconv_t)-1) {
+					iconv(iso8859_1to,0,0,0,0);
+					iconv(iso8859_1to,const_cast<input_buffer_t>(&copy),&len,&dst,&outputLeft);
+					iconv_close(iso8859_1to);
+				}
+				delete original;
+				}
+				break;
+			case EINVAL: // incomplete multibyte sequence in the input
+				// we just eat bad bytes, as part of robustness/best-effort
+				inputBuffer++;
+				inputLeft--;
+				break;
+			case E2BIG:
+				// not enough room in the output buffer for the next converted character
+				break;
+			default:
+				// unknown error, completely bail
+				status = errno;
+				iconv_close(conversion);
+				return status;
+			}
+		}
+	} while ((inputLeft > 0) && (outputLeft > 0));
 	*srcLen -= inputLeft;
 	*dstLen -= outputLeft;
 	iconv_close(conversion);
-	if (nonReversibleConversions == -1) {
-		switch (errno) {
-		case EILSEQ: // invalid multibyte sequence in the source
-			return B_ERROR;
-		case EINVAL: // incomplete multibyte sequence in the input
-			return B_OK;
-		case E2BIG: // not enough room in the output buffer for the next converted character
-			return B_OK;
-		default:
-			// unknown error
-			int err = errno;
-		}
-	}
 	if (*srcLen != 0) {
 		// able to convert at least one character
 		return B_OK;
@@ -68,7 +104,7 @@ convert_to_utf8(uint32 srcEncoding,
 	if (charset == 0) {
 		return B_ERROR;
 	}
-	return convert_encoding(charset->GetName(),"UTF-8",src,srcLen,dst,dstLen,state);
+	return convert_encoding(charset->GetName(),"UTF-8",src,srcLen,dst,dstLen,state,substitute);
 }
 
 status_t
@@ -81,5 +117,5 @@ convert_from_utf8(uint32 dstEncoding,
 	if (charset == 0) {
 		return B_ERROR;
 	}	
-	return convert_encoding("UTF-8",charset->GetName(),src,srcLen,dst,dstLen,state);
+	return convert_encoding("UTF-8",charset->GetName(),src,srcLen,dst,dstLen,state,substitute);
 }