diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/UnivocityParser.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/UnivocityParser.scala
index e42ea3fa391f..b1d7702079ca 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/UnivocityParser.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/UnivocityParser.scala
@@ -254,6 +254,17 @@ class UnivocityParser(
     }
   }
 
+  /**
+   * This function deals with the cases it fails to parse in PERMISSIVE mode. The failure reasons
+   * of this mode are 1) the longer lengths of tokens than expected or 2) format exceptions
+   * (e.g., NumberFormatException).
+   */
+  private def failedRecordWithPermissiveMode(): Option[InternalRow] = {
+    val row = new GenericInternalRow(requiredSchema.length)
+    corruptFieldIndex.foreach(row(_) = UTF8String.fromString(getCurrentInput()))
+    Some(row)
+  }
+
   private def convertWithParseMode(
       tokens: Array[String])(convert: Array[String] => InternalRow): Option[InternalRow] = {
     if (options.dropMalformed && dataSchema.length != tokens.length) {
@@ -271,43 +282,37 @@ class UnivocityParser(
       throw new RuntimeException(s"Malformed line in FAILFAST mode: " +
         s"${tokens.mkString(options.delimiter.toString)}")
     } else {
-      // If a length of parsed tokens is not equal to expected one, it makes the length the same
-      // with the expected. If the length is shorter, it adds extra tokens in the tail.
-      // If longer, it drops extra tokens.
-      //
-      // TODO: Revisit this; if a length of tokens does not match an expected length in the schema,
-      // we probably need to treat it as a malformed record.
-      // See an URL below for related discussions:
-      // https://github.com/apache/spark/pull/16928#discussion_r102657214
-      val checkedTokens = if (options.permissive && dataSchema.length != tokens.length) {
-        if (dataSchema.length > tokens.length) {
+      // If a length of parsed tokens is longer than expected, it treats them as malformed.
+      if (options.permissive && dataSchema.length < tokens.length) {
+        failedRecordWithPermissiveMode()
+      } else {
+        // If the length is shorter than expected, it adds extra tokens in the tail.
+        val checkedTokens = if (options.permissive && dataSchema.length > tokens.length) {
           tokens ++ new Array[String](dataSchema.length - tokens.length)
         } else {
-          tokens.take(dataSchema.length)
+          tokens
         }
-      } else {
-        tokens
-      }
 
-      try {
-        Some(convert(checkedTokens))
-      } catch {
-        case NonFatal(e) if options.permissive =>
-          val row = new GenericInternalRow(requiredSchema.length)
-          corruptFieldIndex.foreach(row(_) = UTF8String.fromString(getCurrentInput()))
-          Some(row)
-        case NonFatal(e) if options.dropMalformed =>
-          if (numMalformedRecords < options.maxMalformedLogPerPartition) {
-            logWarning("Parse exception. " +
-              s"Dropping malformed line: ${tokens.mkString(options.delimiter.toString)}")
-          }
-          if (numMalformedRecords == options.maxMalformedLogPerPartition - 1) {
-            logWarning(
-              s"More than ${options.maxMalformedLogPerPartition} malformed records have been " +
-                "found on this partition. Malformed records from now on will not be logged.")
-          }
-          numMalformedRecords += 1
-          None
+        try {
+          Some(convert(checkedTokens))
+        } catch {
+          case NonFatal(e) if options.permissive =>
+            val row = new GenericInternalRow(requiredSchema.length)
+            corruptFieldIndex.foreach(row(_) = UTF8String.fromString(getCurrentInput()))
+            Some(row)
+          case NonFatal(e) if options.dropMalformed =>
+            if (numMalformedRecords < options.maxMalformedLogPerPartition) {
+              logWarning("Parse exception. " +
+                s"Dropping malformed line: ${tokens.mkString(options.delimiter.toString)}")
+            }
+            if (numMalformedRecords == options.maxMalformedLogPerPartition - 1) {
+              logWarning(
+                s"More than ${options.maxMalformedLogPerPartition} malformed records have been " +
+                  "found on this partition. Malformed records from now on will not be logged.")
+            }
+            numMalformedRecords += 1
+            None
+        }
       }
     }
   }
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVSuite.scala
index 95dfdf5b298e..bbbec937504a 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVSuite.scala
@@ -297,17 +297,6 @@ class CSVSuite extends QueryTest with SharedSQLContext with SQLTestUtils {
     }
   }
 
-  test("test for tokens more than the fields in the schema") {
-    val cars = spark
-      .read
-      .format("csv")
-      .option("header", "false")
-      .option("comment", "~")
-      .load(testFile(carsMalformedFile))
-
-    verifyCars(cars, withHeader = false, checkTypes = false)
-  }
-
   test("test with null quote character") {
     val cars = spark.read
       .format("csv")
@@ -1116,4 +1105,21 @@ class CSVSuite extends QueryTest with SharedSQLContext with SQLTestUtils {
     assert(df2.schema === schema)
   }
 
+  test("SPARK-19783 test for tokens more than the fields in the schema") {
+    val columnNameOfCorruptRecord = "_unparsed"
+    withTempPath { path =>
+      Seq("1,2", "1,2,3,4").toDF().write.text(path.getAbsolutePath)
+      val schema = StructType(
+        StructField("a", IntegerType, true) ::
+        StructField("b", IntegerType, true) ::
+        StructField(columnNameOfCorruptRecord, StringType, true) :: Nil)
+      val df = spark.read
+        .schema(schema)
+        .option("header", "false")
+        .option("columnNameOfCorruptRecord", columnNameOfCorruptRecord)
+        .csv(path.getAbsolutePath)
+
+      checkAnswer(df, Row(1, 2, null) :: Row(null, null, "1,2,3,4") :: Nil)
+    }
+  }
 }