Index: lucene/suggest/src/test/org/apache/lucene/search/suggest/analyzing/AnalyzingSuggesterTest.java
===================================================================
--- lucene/suggest/src/test/org/apache/lucene/search/suggest/analyzing/AnalyzingSuggesterTest.java	(revision 1405894)
+++ lucene/suggest/src/test/org/apache/lucene/search/suggest/analyzing/AnalyzingSuggesterTest.java	(working copy)
@@ -983,4 +983,17 @@
     assertEquals("b", results.get(1).key);
     assertEquals(5, results.get(1).value);
   }
+
+  public void testDupSurfaceFormsMissingResults3() throws Exception {
+    Analyzer a = new MockAnalyzer(random());
+    AnalyzingSuggester suggester = new AnalyzingSuggester(a, a, AnalyzingSuggester.PRESERVE_SEP, 256, -1);
+    suggester.build(new TermFreqArrayIterator(new TermFreq[] {
+          new TermFreq("a a", 7),
+          new TermFreq("a a", 7),
+          new TermFreq("a c", 6),
+          new TermFreq("a c", 3),
+          new TermFreq("a b", 5),
+        }));
+    assertEquals("[a a/7, a c/6, a b/5]", suggester.lookup("a", false, 3).toString());
+  }
 }
Index: lucene/suggest/src/java/org/apache/lucene/search/suggest/analyzing/AnalyzingSuggester.java
===================================================================
--- lucene/suggest/src/java/org/apache/lucene/search/suggest/analyzing/AnalyzingSuggester.java	(revision 1405894)
+++ lucene/suggest/src/java/org/apache/lucene/search/suggest/analyzing/AnalyzingSuggester.java	(working copy)
@@ -340,6 +340,7 @@
     try {
       ByteArrayDataOutput output = new ByteArrayDataOutput(buffer);
       BytesRef surfaceForm;
+
       while ((surfaceForm = iterator.next()) != null) {
         Set<IntsRef> paths = toFiniteStrings(surfaceForm, ts2a);
         
@@ -379,6 +380,10 @@
 
       // Sort all input/output pairs (required by FST.Builder):
       new Sort().sort(tempInput, tempSorted);
+
+      // Free disk space:
+      tempInput.delete();
+
       reader = new Sort.ByteSequencesReader(tempSorted);
      
       PairOutputs<Long,BytesRef> outputs = new PairOutputs<Long,BytesRef>(PositiveIntOutputs.getSingleton(true), ByteSequenceOutputs.getSingleton());
@@ -391,6 +396,12 @@
       IntsRef scratchInts = new IntsRef();
       ByteArrayDataInput input = new ByteArrayDataInput();
 
+      // Used to remove duplicate surface forms (but we
+      // still index the hightest-weight one).  We clear
+      // this when we see a new analyzed form, so it cannot
+      // grow unbounded (at most 256 entries):
+      Set<BytesRef> seenSurfaceForms = new HashSet<BytesRef>();
+
       int dedup = 0;
       while (reader.read(scratch)) {
         input.reset(scratch.bytes, scratch.offset, scratch.length);
@@ -411,6 +422,7 @@
         if (previous == null) {
           previous = new BytesRef();
           previous.copyBytes(analyzed);
+          seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
         } else if (analyzed.equals(previous)) {
           dedup++;
           if (dedup >= maxSurfaceFormsPerAnalyzedForm) {
@@ -418,9 +430,15 @@
             // dups: skip the rest:
             continue;
           }
+          if (seenSurfaceForms.contains(surface)) {
+            continue;
+          }
+          seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
         } else {
           dedup = 0;
           previous.copyBytes(analyzed);
+          seenSurfaceForms.clear();
+          seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
         }
 
         analyzed.grow(analyzed.length+2);