diff --git a/src/main/scala/nak/core/Classifier.scala b/src/main/scala/nak/core/Classifier.scala
index 887e8a5..4930819 100644
--- a/src/main/scala/nak/core/Classifier.scala
+++ b/src/main/scala/nak/core/Classifier.scala
@@ -102,15 +102,17 @@ trait LiblinearClassifier extends IndexedClassifier[String] {
 
   /**
    * Implement the apply method of Classifier by transforming the tuples into
-   * Liblinear Features and then calling Linear.predictProbability.
-   *
-   * TODO: This should be made more general so that the SVM solvers can be used
-   * by Nak.
+   * Liblinear Features and then calling Linear.predictProbability or
+   * Linear.predictValues, depending if the model supports probability or not.
    */ 
   def apply(context: Array[(Int,Double)]): Array[Double] = {
     val ctxt = context.map(c=>new FeatureNode(c._1,c._2).asInstanceOf[LiblinearFeature])
     val labelScores = Array.fill(numLabels)(0.0)
-    Linear.predictProbability(model, ctxt, labelScores)
+    if (model.isProbabilityModel) {
+      Linear.predictProbability(classifier.model, ctxt, labelScores)
+    } else {
+      Linear.predictValues(model, ctxt, labelScores)
+    }
     labelScores
   }