diff --git a/xwords4/dawg/dawg2dict.py b/xwords4/dawg/dawg2dict.py
new file mode 100755
index 000000000..fadfb1f48
--- /dev/null
+++ b/xwords4/dawg/dawg2dict.py
@@ -0,0 +1,250 @@
+#!/usr/bin/env python3
+
+import argparse, io, struct, sys
+
+oneByteFmt = struct.Struct('B')
+SPACE = ' '
+
+def getNullTermParam(fh):
+    msg = ""
+    while True:
+        (oneChar) = oneByteFmt.unpack(fh.read(oneByteFmt.size))
+        if int(oneChar[0]) == 0: break
+        msg += chr(oneChar[0])
+    return msg
+
+def addFace( faces, face ):
+    assert face
+    faces.append( face )
+    # print( 'addFace(): added:', face, ' now have', len(faces), 'faces' )
+
+# Each face is one or two synonym strings (typically the upper- and
+# lower-case versions of a tile face), with a space as separator in the
+# two-string case.
+#
+# Each letter read is a space/delimiter or not. If it's a delimiter,
+# then we append what we had before. Otherwise if we have something
+# already then this is a new face starting. Otherwise it's a synonym
+# for something we're processing already.
+    
+def splitFaces( buf ):
+    faces = []
+    synonyms = None
+    lastWasDelim = False
+    for oneChar in buf:
+        # print('read char', oneChar)
+
+        if oneChar == SPACE:
+            assert synonyms     # there's better be one already
+            lastWasDelim = True
+        else:
+            # print( "read non-delim char:", oneChar )
+            if lastWasDelim:
+                assert len(synonyms) == 1
+                synonyms.append( oneChar )
+                addFace( faces, synonyms )
+                synonyms = None
+                lastWasDelim = False
+            else:
+                if synonyms:
+                    addFace( faces, synonyms )
+                synonyms = [ oneChar ]
+
+    if synonyms: addFace( faces, synonyms )
+
+    return faces
+
+def loadCountsAndValues( fh, numFaces, data ):
+    twoBytesFmt = struct.Struct('BB')
+    for ii in range(numFaces):
+        pair = twoBytesFmt.unpack(fh.read(twoBytesFmt.size))
+        data[ii]['count'] = int(pair[0])
+        data[ii]['val'] = int(pair[1])
+
+def eatBitmap( fh ):
+    nCols = int(oneByteFmt.unpack(fh.read(oneByteFmt.size))[0])
+    if nCols > 0:
+        nRows = int(oneByteFmt.unpack(fh.read(oneByteFmt.size))[0])
+        nBytes = ((nRows*nCols)+7) // 8
+        print('eatBitmap(): skipping', nBytes, 'bytes; nCols:', nCols, 'nRows:', nRows)
+        fh.read(nBytes)
+
+def loadSpecialData( fh, data ):
+    count = 0
+    lastSpecial = ord(' ')
+    for datum in data:
+        # print('loadSpecialData: comparing', ord(datum['faces'][0]), 'with', lastSpecial)
+        if len(datum['faces']) == 1 and ord(datum['faces'][0]) < lastSpecial:
+            count += 1
+            txtlen = int(oneByteFmt.unpack(fh.read(oneByteFmt.size))[0])
+            txt = fh.read(txtlen).decode("UTF-8")
+            # print('loadSpecialData(): found:', txt, 'of len', txtlen)
+            datum['faces'] = txt.split( SPACE )
+            eatBitmap( fh )
+            eatBitmap( fh )
+
+def loadNodes( dawg, nodeSize ):
+    nodes = []
+    fmtStr = ''
+    for ii in range(nodeSize): fmtStr += 'B'
+    fmt = struct.Struct(fmtStr)
+
+    while True:
+        buf = dawg.read(nodeSize)
+        if len(buf) == 0: break
+        assert len(buf) == nodeSize
+        arr = fmt.unpack(buf)
+        val = 0
+        for elem in arr:
+            val = (val << 8) + elem
+        nodes.append(val)
+        # print('loaded node 0x{:x} of len {}'.format(val, nodeSize))
+    return nodes
+
+def parseNode( node, nodeSize ):
+
+    if nodeSize == 4:
+        accepting = (node & 0x00008000) != 0
+        isLast = (node & 0x00004000) != 0
+        chrIndex = (node & 0x00003f00) >> 8
+        nextEdge = (node >> 16) + ((node & 0x000000FF) << 16)
+    elif nodeSize == 3:
+        accepting = (node & 0x00000080) != 0
+        isLast = (node & 0x00000040) != 0
+        chrIndex = node & 0x0000001f
+        nextEdge = (node >> 8) + ((node & 0x00000020) << 11)
+    
+    return (nextEdge, chrIndex, accepting, isLast )
+
+def expandDAWG( nodes, nodeSize, indx, data, words, letters = [] ):
+    if len(letters) > 15: error( "infinite recursion???" )
+
+    while True:
+        node = nodes[indx]
+        indx += 1
+        ( nextEdge, chrIndex, accepting, lastEdge ) = parseNode( node, nodeSize )
+
+        letters.append( data[chrIndex]['faces'][0] )
+        if accepting:
+            words.append( ''.join(letters) )
+
+        if nextEdge != 0:
+            expandDAWG( nodes, nodeSize, nextEdge, data, words, letters )
+
+        letters.pop()
+
+        if lastEdge: break
+
+def process(args):
+    DICT_SYNONYMS_MASK = 0x10
+    DICT_HEADER_MASK = 0x08
+
+    with open(args.DAWG, "rb") as dawg:
+        nWords = 0
+
+        headerFmt = struct.Struct('!HH')
+        (flags, headerLen) = headerFmt.unpack(dawg.read(headerFmt.size))
+        print( 'read flags: {:x}, header len: {}'.format(flags, headerLen ),
+               file=sys.stderr )
+        if not 0 == DICT_HEADER_MASK & flags:
+            flags &= ~DICT_HEADER_MASK
+            header = io.BytesIO(dawg.read(headerLen))
+            nWordsFmt = struct.Struct('!L')
+            nWords = nWordsFmt.unpack(header.read(nWordsFmt.size))[0]
+            
+            print( 'header: read nWords: {}'.format(nWords ), file=sys.stderr )
+
+            msg = getNullTermParam(header)
+            if args.DUMP_MSG:
+                print( 'msg: {}'.format(msg))
+            md5Sum = getNullTermParam(header)
+            print( 'header: read sum: {}'.format(md5Sum), file=sys.stderr )
+
+        nodeSize = 0
+        isUTF8 = False
+        flags &= ~DICT_SYNONYMS_MASK
+        if flags == 0x0002:
+            nodeSize = 3
+        elif flags == 0x0003:
+            nodeSize = 4
+        elif flags == 0x0004:
+            isUTF8 = True
+            nodeSize = 3
+        elif flags == 0x0005:
+            isUTF8 = True
+            nodeSize = 4
+        else:
+            error("unexpected flags value")
+        print( 'nodesize: {}, isUTF8: {}'.format(nodeSize, isUTF8), file=sys.stderr )
+
+        numFaceBytes = 0
+        if isUTF8:
+            numFaceBytes = oneByteFmt.unpack(dawg.read(oneByteFmt.size))[0]
+        numFaces = int(oneByteFmt.unpack(dawg.read(oneByteFmt.size))[0])
+        if not isUTF8:
+            numFaceBytes = numFaces * 2
+        if numFaces > 64:
+            error("too many faces: " + numFaces)
+        print( 'numFaceBytes: {}, numFaces: {}'.format(numFaceBytes, numFaces), file=sys.stderr )
+
+        print( 'TODO: confirm checksum', file=sys.stderr )
+
+        data = []
+        if isUTF8:
+            faceBytes = dawg.read(numFaceBytes).decode("UTF-8")
+            faces = splitFaces( faceBytes )
+            assert( len(faces) == numFaces )
+            # print( 'loaded', len(faces), 'faces:', faces )
+            for datum in faces:
+                data.append({'faces' : datum })
+        else:
+            error('I don\'t handle obsolete ascii case')
+
+        langCode = 0x7F & oneByteFmt.unpack(dawg.read(oneByteFmt.size))[0]
+        dawg.read( oneByteFmt.size ) # skip byte
+
+        loadCountsAndValues( dawg, numFaces, data )
+        loadSpecialData( dawg, data )
+
+        offsetStruct = struct.Struct('!L')
+        assert offsetStruct.size == 4
+        offset = int(offsetStruct.unpack(dawg.read(offsetStruct.size))[0])
+
+        if args.DUMP_TILES:
+            for ii in range(len(data)):
+                print( 'tile {:2d}: {}:'.format(ii, data[ii]) )
+
+        nodes = loadNodes( dawg, nodeSize )
+        words = []
+        expandDAWG( nodes, nodeSize, offset, data, words )
+        assert len(words) == nWords
+        if args.DUMP_WORDS:
+            for word in words:
+                print(word)
+
+def mkParser():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--dict', dest = 'DAWG', type = str, required = True,
+                        help = 'the .xwd file to load')
+    parser.add_argument('--dump-words', dest = 'DUMP_WORDS', default = False,
+                        action = 'store_true', help = 'write wordlist to stdout')
+    parser.add_argument('--dump-tiles', dest = 'DUMP_TILES', default = False,
+                        action = 'store_true', help = 'write tile metadata to stdout')
+    parser.add_argument('--dump-msg', dest = 'DUMP_MSG', default = False,
+                        action = 'store_true', help = 'write header user-visible message to stdout')
+
+
+    # [-raw | -json]  [-get-sum] [-get-desc] -dict <xwdORpdb>
+
+    return parser
+
+def parseArgs():
+    args = mkParser().parse_args()
+    process( args )
+
+def main():
+    args = parseArgs()
+
+##############################################################################
+if __name__ == '__main__':
+    main()