diff --git a/src/bayesoptbase.cpp b/src/bayesoptbase.cpp
index ac9edad..10f4067 100644
--- a/src/bayesoptbase.cpp
+++ b/src/bayesoptbase.cpp
@@ -144,6 +144,10 @@ namespace bayesopt
     
     mModel->addSample(xNext,yNext);
 
+    // Update surrogate model
+    bool retrain = ((mParameters.n_iter_relearn > 0) && 
+		    ((mCurrentIter + 1) % mParameters.n_iter_relearn == 0));
+
     if (mUseRobust)
       {
         mFilter->addSample(xNext,yNext);
@@ -156,13 +160,11 @@ namespace bayesopt
           {
             //TODO: Check that the copy is safe.
             mModel->copyData(mFilter->filterPoints());
+            retrain = true;
           }
         
       }
     
-    // Update surrogate model
-    bool retrain = ((mParameters.n_iter_relearn > 0) && 
-		    ((mCurrentIter + 1) % mParameters.n_iter_relearn == 0));
 
     if (retrain)  // Full update
       {
diff --git a/src/robust_filtering.cpp b/src/robust_filtering.cpp
index ea8595e..b7d646b 100644
--- a/src/robust_filtering.cpp
+++ b/src/robust_filtering.cpp
@@ -37,12 +37,15 @@ namespace bayesopt
   
   const Dataset* RobustFiltering::filterPoints()
   {
-    mFilteredData.reset();
+    mFilteredData.reset(new Dataset());
     vecOfvec XX = mRobustModel->getData()->mX;
     vectord YY = mRobustModel->getData()->mY;
 
     size_t n_points = mRobustModel->getData()->getNSamples();
 
+    mRobustModel->updateHyperParameters();
+    mRobustModel->fitSurrogateModel();
+
     for(size_t i = 0; i < n_points; ++i)
       {
         ProbabilityDistribution* pd = mRobustModel->getPrediction(XX[i]);
@@ -54,6 +57,12 @@ namespace bayesopt
             utils::append(mFilteredData->mY, YY[i]);
           }
       }
+    if (mFilteredData->getNSamples() <= n_points * 0.5)
+      {
+        mFilteredData->mX = XX;
+        mFilteredData->mY = YY;
+      }
+    
     return mFilteredData.get();
   }