diff --git a/svm.c b/svm.c
index 983084a..b67cbe8 100644
--- a/svm.c
+++ b/svm.c
@@ -970,21 +970,23 @@ PHP_METHOD(svmmodel, predict)
 }
 /* }}} */
 
-/** {{{ SvmModel::predict_probability(array data)
+/** {{{ SvmModel::predict_probability(array data, array probabilities)
 	Predicts based on the model
 */
 PHP_METHOD(svmmodel, predict_probability)
 {
 	php_svm_model_object *intern;
 	double predict_probability;
-	int nr_classes;
+	int nr_classes, i;
 	double *estimates;
 	struct svm_node *x;
 	int max_nr_attr = 64;
-	zval *arr;
+	int *labels;
+	zval *arr; 
+	zval *retarr = NULL;
 	
 	/* we want an array of data to be passed in */
-	if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "a", &arr) == FAILURE) {
+	if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "az", &arr, &retarr) == FAILURE) {
 	    return;
 	}
 
@@ -996,9 +998,20 @@ PHP_METHOD(svmmodel, predict_probability)
 	x = php_svm_get_data_from_array(arr TSRMLS_CC);
 	nr_classes = svm_get_nr_class(intern->model);
 	estimates = safe_emalloc(nr_classes, sizeof(double), 0);
+	labels = safe_emalloc(nr_classes, sizeof(int), 0);
 	predict_probability = svm_predict_probability(intern->model, x, estimates);
-	predict_probability = estimates[(int)predict_probability];
+	
+	if (retarr != NULL) {
+		zval_dtor(retarr);
+		array_init(retarr);
+		svm_get_labels(intern->model, labels);
+		for (i = 0; i < nr_classes; ++i) {
+			add_index_double(retarr, labels[i], estimates[i]);
+		}
+	}
+	
 	efree(estimates);
+	efree(labels);
 	efree(x);
 	
 	RETURN_DOUBLE(predict_probability);
diff --git a/tests/014_predict_probability.phpt b/tests/014_predict_probability.phpt
index 9837679..7630bdf 100644
--- a/tests/014_predict_probability.phpt
+++ b/tests/014_predict_probability.phpt
@@ -11,7 +11,7 @@ $svm->setOptions(array(
 	SVM::OPT_TYPE => SVM::C_SVC,
 	SVM::OPT_KERNEL_TYPE => SVM::KERNEL_LINEAR,
 	SVM::OPT_P => 0.1,  // epsilon 0.1
-	SVM::OPT_PROBABILITY => true
+	SVM::OPT_PROBABILITY => 1
 ));
 $model = $svm->train(dirname(__FILE__) . '/abalone.scale');
 
@@ -27,8 +27,12 @@ if($model) {
 		8 => -0.704036
 	);
 	$class = $model->predict($data);
-	$result = $model->predict_probability($data);
-	if($class == 9 && $result > 0) {
+	$return = array();
+	$result = $model->predict_probability($data, $return);
+    arsort($return);
+    reset($return);
+    $key = key($return);
+	if($class == 9 && $key == 9) {
 		echo "ok";
 	} else {
 		echo "predict failed: $class $result";