diff --git a/backends-clickhouse/src/main/scala/org/apache/gluten/backendsapi/clickhouse/CHSparkPlanExecApi.scala b/backends-clickhouse/src/main/scala/org/apache/gluten/backendsapi/clickhouse/CHSparkPlanExecApi.scala
index ba165d936eed..e17e15b04bd7 100644
--- a/backends-clickhouse/src/main/scala/org/apache/gluten/backendsapi/clickhouse/CHSparkPlanExecApi.scala
+++ b/backends-clickhouse/src/main/scala/org/apache/gluten/backendsapi/clickhouse/CHSparkPlanExecApi.scala
@@ -158,16 +158,38 @@ class CHSparkPlanExecApi extends SparkPlanExecApi with Logging {
       aggregateAttributes: Seq[Attribute],
       initialInputBufferOffset: Int,
       resultExpressions: Seq[NamedExpression],
-      child: SparkPlan): HashAggregateExecBaseTransformer =
-    CHHashAggregateExecTransformer(
+      child: SparkPlan): HashAggregateExecBaseTransformer = {
+    logError(s"xxx aggregateExpressions:$aggregateExpressions")
+    logError(s"xxx aggregateAttributes:$aggregateAttributes")
+    logError(s"xxx resultExpressions:$resultExpressions")
+    logError(s"xxx agg expr to result: ${aggregateExpressions.map(_.resultAttribute)}")
+    logError(
+      s"xxx agg:" +
+        s"${aggregateExpressions.map(e => e.aggregateFunction.aggBufferAttributes.length)}")
+    aggregateExpressions.foreach {
+      e => logError(s"xxx agg fun:$e, ${e.aggregateFunction.aggBufferAttributes}")
+    }
+    val modes = aggregateExpressions.map(_.mode)
+    logError(s"xxx modes: $modes")
+    val xoutputs = CHHashAggregateExecTransformer.getCHAggregateResultAttributes(
+      aggregateExpressions,
+      resultExpressions.slice(groupingExpressions.length, resultExpressions.length))
+    logError(s"xxx adjust agg output: $xoutputs")
+    val replacedResultExpressions =
+      groupingExpressions ++ xoutputs
+    val agg = CHHashAggregateExecTransformer(
       requiredChildDistributionExpressions,
       groupingExpressions.distinct,
       aggregateExpressions,
       aggregateAttributes,
       initialInputBufferOffset,
-      resultExpressions.distinct,
+      // resultExpressions.distinct,
+      replacedResultExpressions,
       child
     )
+    logError(s"xxx agg output: ${agg.output}")
+    agg
+  }
 
   /** Generate HashAggregateExecPullOutHelper */
   override def genHashAggregateExecPullOutHelper(
diff --git a/backends-clickhouse/src/main/scala/org/apache/gluten/execution/CHHashAggregateExecTransformer.scala b/backends-clickhouse/src/main/scala/org/apache/gluten/execution/CHHashAggregateExecTransformer.scala
index f5e64330cd15..8a0b66fcf191 100644
--- a/backends-clickhouse/src/main/scala/org/apache/gluten/execution/CHHashAggregateExecTransformer.scala
+++ b/backends-clickhouse/src/main/scala/org/apache/gluten/execution/CHHashAggregateExecTransformer.scala
@@ -43,6 +43,42 @@ import scala.collection.JavaConverters._
 import scala.collection.mutable.ListBuffer
 
 object CHHashAggregateExecTransformer {
+  // The result attributes of aggregate expressions from vanilla may be different from CH native.
+  // For example, the result attributes of `avg(x)` are `sum(x)` and `count(x)`. This could bring
+  // some unexpected issues. So we need to make the result attributes consistent with CH native.
+  def getCHAggregateResultAttributes(
+      aggregateExpressions: Seq[AggregateExpression],
+      resultExpressions: Seq[NamedExpression]): Seq[Attribute] = {
+    var resultExpressionIndex = 0
+    aggregateExpressions.flatMap {
+      aggExpr =>
+        aggExpr.mode match {
+          case Partial | PartialMerge =>
+            val aggBufferAttributesCount = aggExpr.aggregateFunction.aggBufferAttributes.length
+            aggExpr.aggregateFunction match {
+              case avg: Average =>
+                val res = Seq(aggExpr.resultAttribute)
+                resultExpressionIndex += aggBufferAttributesCount
+                res
+              case sum: Sum if (sum.dataType.isInstanceOf[DecimalType]) =>
+                val res = Seq(resultExpressions(resultExpressionIndex).toAttribute)
+                resultExpressionIndex += aggBufferAttributesCount
+                res
+              case _ =>
+                val res = resultExpressions
+                  .slice(resultExpressionIndex, resultExpressionIndex + aggBufferAttributesCount)
+                  .map(_.toAttribute)
+                resultExpressionIndex += aggBufferAttributesCount
+                res
+            }
+          case _ =>
+            val res = Seq(resultExpressions(resultExpressionIndex).toAttribute)
+            resultExpressionIndex += 1
+            res
+        }
+    }
+  }
+
   def getAggregateResultAttributes(
       groupingExpressions: Seq[NamedExpression],
       aggregateExpressions: Seq[AggregateExpression]): Seq[Attribute] = {