diff --git a/python/pyspark/sql/tests.py b/python/pyspark/sql/tests.py
index 8d738069adb3d..4404dbe40590a 100644
--- a/python/pyspark/sql/tests.py
+++ b/python/pyspark/sql/tests.py
@@ -5925,6 +5925,22 @@ def test_invalid_args(self):
                     'mixture.*aggregate function.*group aggregate pandas UDF'):
                 df.groupby(df.id).agg(mean_udf(df.v), mean(df.v)).collect()
 
+    def test_self_join_with_pandas(self):
+        import pyspark.sql.functions as F
+
+        @F.pandas_udf('key long, col string', F.PandasUDFType.GROUPED_MAP)
+        def dummy_pandas_udf(df):
+            return df[['key', 'col']]
+
+        df = self.spark.createDataFrame([Row(key=1, col='A'), Row(key=1, col='B'),
+                                         Row(key=2, col='C')])
+        dfWithPandas = df.groupBy('key').apply(dummy_pandas_udf)
+
+        # this was throwing an AnalysisException before SPARK-24208
+        res = dfWithPandas.alias('temp0').join(dfWithPandas.alias('temp1'),
+                                               F.col('temp0.key') == F.col('temp1.key'))
+        self.assertEquals(res.count(), 5)
+
 
 @unittest.skipIf(
     not _have_pandas or not _have_pyarrow,
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
index e187133d03b17..c078efdfc0000 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
@@ -738,6 +738,10 @@ class Analyzer(
             if findAliases(aggregateExpressions).intersect(conflictingAttributes).nonEmpty =>
           (oldVersion, oldVersion.copy(aggregateExpressions = newAliases(aggregateExpressions)))
 
+        case oldVersion @ FlatMapGroupsInPandas(_, _, output, _)
+            if oldVersion.outputSet.intersect(conflictingAttributes).nonEmpty =>
+          (oldVersion, oldVersion.copy(output = output.map(_.newInstance())))
+
         case oldVersion: Generate
             if oldVersion.producedAttributes.intersect(conflictingAttributes).nonEmpty =>
           val newOutput = oldVersion.generatorOutput.map(_.newInstance())
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/GroupedDatasetSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/GroupedDatasetSuite.scala
index 147c0b61f5017..bd54ea415ca88 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/GroupedDatasetSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/GroupedDatasetSuite.scala
@@ -93,4 +93,16 @@ class GroupedDatasetSuite extends QueryTest with SharedSQLContext {
     }
     datasetWithUDF.unpersist(true)
   }
+
+  test("SPARK-24208: analysis fails on self-join with FlatMapGroupsInPandas") {
+    val df = datasetWithUDF.groupBy("s").flatMapGroupsInPandas(PythonUDF(
+      "pyUDF",
+      null,
+      StructType(Seq(StructField("s", LongType))),
+      Seq.empty,
+      PythonEvalType.SQL_GROUPED_MAP_PANDAS_UDF,
+      true))
+    val df1 = df.alias("temp0").join(df.alias("temp1"), $"temp0.s" === $"temp1.s")
+    df1.queryExecution.assertAnalyzed()
+  }
 }