Matched sampling is a methodology used to estimate treatment effects. A caliper mechanism is used to achieve better similarity among matched pairs. We investigate finite sample properties of matching with caliper and propose a slight modification to the existing mechanism. The simulation study compare performance of both methods and show that standard caliper perform well only in case of constant treatment or uniform propensity score distribution. Secondly, in a case of non-uniform distribution and non-uniform treatment the dynamic caliper method outperform standard caliper matching.