diff --git a/RansacLib/ransac.h b/RansacLib/ransac.h
index a2f3637..516de95 100644
--- a/RansacLib/ransac.h
+++ b/RansacLib/ransac.h
@@ -259,7 +259,8 @@ class LocallyOptimizedMSAC : public RansacBase {
       solver.LeastSquares(stats.inlier_indices, &refined_model);
 
       double score = std::numeric_limits<double>::max();
-      ScoreModel(solver, refined_model, kSqrInlierThresh, &score);
+      ScoreModel(solver, refined_model, kSqrInlierThresh, &score, 
+                 stats.best_model_score);
       if (score < stats.best_model_score) {
         stats.best_model_score = score;
         *best_model = refined_model;
@@ -283,7 +284,8 @@ class LocallyOptimizedMSAC : public RansacBase {
     *best_model_id = 0;
     for (int m = 0; m < num_models; ++m) {
       double score = std::numeric_limits<double>::max();
-      ScoreModel(solver, models[m], squared_inlier_threshold, &score);
+      ScoreModel(solver, models[m], squared_inlier_threshold, &score,
+                 *best_score);
 
       if (score < *best_score) {
         *best_score = score;
@@ -293,12 +295,16 @@ class LocallyOptimizedMSAC : public RansacBase {
   }
 
   void ScoreModel(const Solver& solver, const Model& model,
-                  const double squared_inlier_threshold, double* score) const {
+                  const double squared_inlier_threshold, double* score
+                  const double best_score) const {
     const int kNumData = solver.num_data();
     *score = 0.0;
     for (int i = 0; i < kNumData; ++i) {
       double squared_error = solver.EvaluateModelOnPoint(model, i);
       *score += ComputeScore(squared_error, squared_inlier_threshold);
+      if( *score > best_score) {
+        return;
+      }
     }
   }
 
@@ -361,7 +367,7 @@ class LocallyOptimizedMSAC : public RansacBase {
     LeastSquaresFit(options, kSqInThresh * kThreshMult, solver, rng, &m_init);
 
     double score = std::numeric_limits<double>::max();
-    ScoreModel(solver, m_init, kSqInThresh, &score);
+    ScoreModel(solver, m_init, kSqInThresh, &score, score_best_minimal_model);
     UpdateBestModel(score, m_init, score_best_minimal_model,
                     best_minimal_model);
 
@@ -383,7 +389,8 @@ class LocallyOptimizedMSAC : public RansacBase {
       Model m_non_min;
       if (!solver.NonMinimalSolver(sample, &m_non_min)) continue;
 
-      ScoreModel(solver, m_non_min, kSqInThresh, &score);
+      ScoreModel(solver, m_non_min, kSqInThresh, &score, 
+                 score_best_minimal_model);
       UpdateBestModel(score, m_non_min, score_best_minimal_model,
                       best_minimal_model);
 
@@ -398,7 +405,8 @@ class LocallyOptimizedMSAC : public RansacBase {
       for (int i = 0; i < options.num_lsq_iterations_; ++i) {
         LeastSquaresFit(options, thresh, solver, rng, &m_non_min);
 
-        ScoreModel(solver, m_non_min, kSqInThresh, &score);
+        ScoreModel(solver, m_non_min, kSqInThresh, &score,
+                   score_best_minimal_model);
         UpdateBestModel(score, m_non_min, score_best_minimal_model,
                         best_minimal_model);
         thresh -= thresh_mult_update;