diff --git a/registration/include/pcl/registration/impl/registration.hpp b/registration/include/pcl/registration/impl/registration.hpp
index b246958927e..480958b72e8 100644
--- a/registration/include/pcl/registration/impl/registration.hpp
+++ b/registration/include/pcl/registration/impl/registration.hpp
@@ -86,32 +86,33 @@ Registration<PointSource, PointTarget, Scalar>::initCompute()
     target_cloud_updated_ = false;
   }
 
+  if (!PCLBase<PointSource>::initCompute())
+    return (false);
+
   // Update the correspondence estimation
   if (correspondence_estimation_) {
     correspondence_estimation_->setSearchMethodTarget(tree_, force_no_recompute_);
     correspondence_estimation_->setSearchMethodSource(tree_reciprocal_,
                                                       force_no_recompute_reciprocal_);
+    correspondence_estimation_->setIndicesSource(indices_);
   }
 
   // Note: we /cannot/ update the search method on all correspondence rejectors, because
   // we know nothing about them. If they should be cached, they must be cached
   // individually.
 
-  return (PCLBase<PointSource>::initCompute());
+  return (true);
 }
 
 template <typename PointSource, typename PointTarget, typename Scalar>
 bool
 Registration<PointSource, PointTarget, Scalar>::initComputeReciprocal()
 {
-  if (!input_) {
-    PCL_ERROR("[pcl::registration::%s::compute] No input source dataset was given!\n",
-              getClassName().c_str());
+  if (!PCLBase<PointSource>::initCompute())
     return (false);
-  }
 
   if (source_cloud_updated_ && !force_no_recompute_reciprocal_) {
-    tree_reciprocal_->setInputCloud(input_);
+    tree_reciprocal_->setInputCloud(input_, indices_);
     source_cloud_updated_ = false;
   }
   return (true);
diff --git a/test/registration/test_registration.cpp b/test/registration/test_registration.cpp
index 0eda177b8a2..6dc8957df63 100644
--- a/test/registration/test_registration.cpp
+++ b/test/registration/test_registration.cpp
@@ -231,6 +231,45 @@ TEST (PCL, IterativeClosestPoint)
   EXPECT_EQ (transformation (3, 3), 1);
 }
 
+TEST (PCL, IterativeClosestPoint_setIndices)
+{
+  pcl::IndicesPtr indices (new pcl::Indices);
+  for (std::size_t i = 0; i < cloud_source.size(); i += 2)
+    indices->push_back(i);
+
+  IterativeClosestPoint<PointXYZ, PointXYZ> reg_indices;
+  PointCloud<PointXYZ>::ConstPtr source (cloud_source.makeShared());
+  reg_indices.setInputSource(source);
+  reg_indices.setInputTarget(cloud_target.makeShared());
+  reg_indices.setIndices(indices);
+  reg_indices.setMaximumIterations(50);
+  reg_indices.setTransformationEpsilon(1e-8);
+  reg_indices.setMaxCorrespondenceDistance(0.05);
+
+  PointCloud<PointXYZ> cloud_reg_indices;
+  reg_indices.align(cloud_reg_indices);
+
+  PointCloud<PointXYZ>::Ptr source_cropped (new PointCloud<PointXYZ>);
+  pcl::copyPointCloud(*source, *indices, *source_cropped);
+
+  IterativeClosestPoint<PointXYZ, PointXYZ> reg_cropped;
+  reg_cropped.setInputSource(source_cropped);
+  reg_cropped.setInputTarget(cloud_target.makeShared());
+  reg_cropped.setMaximumIterations(50);
+  reg_cropped.setTransformationEpsilon(1e-8);
+  reg_cropped.setMaxCorrespondenceDistance(0.05);
+
+  PointCloud<PointXYZ> cloud_reg_cropped;
+  reg_cropped.align(cloud_reg_cropped);
+
+  Eigen::Matrix4f trans_indices = reg_indices.getFinalTransformation();
+  Eigen::Matrix4f trans_cropped = reg_cropped.getFinalTransformation();
+
+  for (int y = 0; y < 4; y++)
+    for (int x = 0; x < 4; x++)
+      EXPECT_NEAR(trans_indices(y, x), trans_cropped(y, x), 1e-5);
+}
+
 TEST (PCL, IterativeClosestPointWithNormals)
 {
   IterativeClosestPointWithNormals<PointNormal, PointNormal, float> reg_float;