diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVOptions.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVOptions.scala
index a13a5a34b4a84..29e76875ad9a1 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVOptions.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVOptions.scala
@@ -152,7 +152,11 @@ class CSVOptions(
     writerSettings.setIgnoreLeadingWhitespaces(ignoreLeadingWhiteSpaceFlagInWrite)
     writerSettings.setIgnoreTrailingWhitespaces(ignoreTrailingWhiteSpaceFlagInWrite)
     writerSettings.setNullValue(nullValue)
-    writerSettings.setEmptyValue(nullValue)
+    // The Univocity parser parses empty strings as `null` by default. This is the default behavior
+    // for Spark too, since `nullValue` defaults to an empty string and has a higher precedence to
+    // setEmptyValue(). But when `nullValue` is set to a different value, that would mean that the
+    // empty string should be parsed not as `null` but as an empty string.
+    writerSettings.setEmptyValue("")
     writerSettings.setSkipEmptyLines(true)
     writerSettings.setQuoteAllFields(quoteAll)
     writerSettings.setQuoteEscapingEnabled(escapeQuotes)
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 4fe45420b4e77..140ac9b52e640 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
@@ -1248,4 +1248,50 @@ class CSVSuite extends QueryTest with SharedSQLContext with SQLTestUtils {
       Row("0,2013-111-11 12:13:14") :: Row(null) :: Nil
     )
   }
+
+  test("SPARK-17916: An empty string should not be coerced to null when nullValue is passed.") {
+    val litNull: String = null
+    val df = Seq(
+      (1, "John Doe"),
+      (2, ""),
+      (3, "-"),
+      (4, litNull)
+    ).toDF("id", "name")
+
+    // Checks for new behavior where an empty string is not coerced to null when `nullValue` is
+    // set to anything but an empty string literal.
+    withTempPath { path =>
+      df.write
+        .option("nullValue", "-")
+        .csv(path.getAbsolutePath)
+      val computed = spark.read
+        .option("nullValue", "-")
+        .schema(df.schema)
+        .csv(path.getAbsolutePath)
+      val expected = Seq(
+        (1, "John Doe"),
+        (2, ""),
+        (3, litNull),
+        (4, litNull)
+      ).toDF("id", "name")
+
+      checkAnswer(computed, expected)
+    }
+    // Keeps the old behavior where empty string us coerced to nullValue is not passed.
+    withTempPath { path =>
+      df.write
+        .csv(path.getAbsolutePath)
+      val computed = spark.read
+        .schema(df.schema)
+        .csv(path.getAbsolutePath)
+      val expected = Seq(
+        (1, "John Doe"),
+        (2, litNull),
+        (3, "-"),
+        (4, litNull)
+      ).toDF("id", "name")
+
+      checkAnswer(computed, expected)
+    }
+  }
 }