diff --git a/spark/src/main/scala/org/apache/spark/sql/comet/execution/arrow/SparkColumnarArrowReader.scala b/spark/src/main/scala/org/apache/spark/sql/comet/execution/arrow/SparkColumnarArrowReader.scala
index 0af940fab3..157aca7423 100644
--- a/spark/src/main/scala/org/apache/spark/sql/comet/execution/arrow/SparkColumnarArrowReader.scala
+++ b/spark/src/main/scala/org/apache/spark/sql/comet/execution/arrow/SparkColumnarArrowReader.scala
@@ -91,6 +91,11 @@ private[comet] class SparkColumnarArrowReader(
     rowsConsumedInCurrent += rowsToProduce
 
     writer.finish()
+    // ArrowWriter derives the root row count from its per-column writes, so a zero-column
+    // input batch (Spark's count-from-metadata scan: numRows > 0, numCols == 0) would otherwise
+    // produce a root with rowCount == 0 and silently drop the rows. Set rowCount explicitly so
+    // downstream aggregations (e.g. df.count()) see the correct value.
+    getVectorSchemaRoot.setRowCount(rowsToProduce)
     onConversionNs(System.nanoTime() - startNs)
     true
   }
diff --git a/spark/src/test/scala/org/apache/comet/exec/CometExecSuite.scala b/spark/src/test/scala/org/apache/comet/exec/CometExecSuite.scala
index dce303b8c1..67a1920a6b 100644
--- a/spark/src/test/scala/org/apache/comet/exec/CometExecSuite.scala
+++ b/spark/src/test/scala/org/apache/comet/exec/CometExecSuite.scala
@@ -3711,6 +3711,45 @@ class CometExecSuite extends CometTestBase {
     }
   }
 
+  test("SparkToColumnar preserves row count for zero-column input batches (df.count())") {
+    // Regression test: when spark.comet.scan is disabled but convert.parquet is enabled,
+    // Spark's count-from-metadata optimization emits ColumnarBatches with numRows > 0 and
+    // numCols == 0. SparkColumnarArrowReader used to derive the Arrow row count from
+    // per-column writes only, silently producing zero-row Arrow batches in this case and
+    // making df.count() return 0.
+    withSQLConf(
+      SQLConf.USE_V1_SOURCE_LIST.key -> "parquet",
+      CometConf.COMET_NATIVE_SCAN_ENABLED.key -> "false",
+      CometConf.COMET_SPARK_TO_ARROW_ENABLED.key -> "true",
+      CometConf.COMET_CONVERT_FROM_PARQUET_ENABLED.key -> "true") {
+      withTempPath { dir =>
+        val expected = 10000L
+        spark
+          .range(expected)
+          .selectExpr("id as key", "id % 8 as value")
+          .toDF("key", "value")
+          .write
+          .parquet(dir.toString)
+
+        val df = spark.read.parquet(dir.toString)
+        // Materialize the count query explicitly so we can inspect the plan
+        // that runs — Dataset.count() executes groupBy().count() internally
+        // but doesn't expose that plan
+        val countDf = df.groupBy().count()
+        assert(
+          countDf.collect().head.getLong(0) == expected,
+          "df.count() should match number of written rows")
+
+        // Ensure this test actually exercises the SparkColumnarArrowReader code path
+        // guarded by the fix, so future regressions are caught.
+        val sparkToColumnar = collect(countDf.queryExecution.executedPlan) {
+          case s: CometSparkToColumnarExec => s
+        }
+        assert(sparkToColumnar.nonEmpty, "Expected CometSparkToColumnarExec in the executed plan")
+      }
+    }
+  }
+
   test("read CSV file") {
     Seq("", "csv").foreach { v1List =>
       withSQLConf(