diff --git a/doc/source/servers/mlflow.md b/doc/source/servers/mlflow.md
index 8b3fbb8760..672b6ab241 100644
--- a/doc/source/servers/mlflow.md
+++ b/doc/source/servers/mlflow.md
@@ -52,6 +52,32 @@ spec:
       replicas: 1
 ```
 
+## MLFlow xtype
+
+By default the server will call your loaded model's predict function with a `numpy.ndarray`. If you wish for it to call it with `pandas.DataFrame` instead, you can pass a parameter `X_type` and set it to `DataFrame`. For example:   
+
+```yaml
+apiVersion: machinelearning.seldon.io/v1alpha2
+kind: SeldonDeployment
+metadata:
+  name: mlflow
+spec:
+  name: wines
+  predictors:
+    - graph:
+        children: []
+        implementation: MLFLOW_SERVER
+        modelUri: gs://seldon-models/mlflow/elasticnet_wine
+        name: classifier
+      parameters:
+        - name: xtype
+          type: STRING
+          value: DataFrame
+      name: default
+      replicas: 1
+```
+
+```
 You can also try out a [worked
 notebook](../examples/server_examples.html#Serve-MLflow-Elasticnet-Wines-Model)
 or check our [talk at the Spark + AI Summit
diff --git a/servers/mlflowserver/mlflowserver/MLFlowServer.py b/servers/mlflowserver/mlflowserver/MLFlowServer.py
index 85f2fcd666..44f8337c0d 100644
--- a/servers/mlflowserver/mlflowserver/MLFlowServer.py
+++ b/servers/mlflowserver/mlflowserver/MLFlowServer.py
@@ -15,10 +15,12 @@
 
 
 class MLFlowServer(SeldonComponent):
-    def __init__(self, model_uri: str):
+    def __init__(self, model_uri: str, xtype: str = 'ndarray'):
         super().__init__()
         logger.info(f"Creating MLFLow server with URI {model_uri}")
+        logger.info(f"xtype: {xtype}")
         self.model_uri = model_uri
+        self.xtype = xtype
         self.ready = False
 
     def load(self):
@@ -34,11 +36,15 @@ def predict(
 
         if not self.ready:
             raise requests.HTTPError("Model not loaded yet")
-        if feature_names is not None and len(feature_names) > 0:
-            df = pd.DataFrame(data=X, columns=feature_names)
+
+        if self.xtype == "ndarray":
+            result = self._model.predict(X)
         else:
-            df = pd.DataFrame(data=X)
-        result = self._model.predict(df)
+            if feature_names is not None and len(feature_names) > 0:
+                df = pd.DataFrame(data=X, columns=feature_names)
+            else:
+                df = pd.DataFrame(data=X)
+            result = self._model.predict(df)
         logger.info(f"Prediction result: {result}")
         return result