Methods for performing motion estimation in multiple reference frames scenario are disclosed. The method includes considering a motion vector of a plurality of neighbouring macro blocks present in a current frame and a co-located frame to find a best prediction point motion vector set, scaling at least one motion vector in a plurality of prediction frames and a plurality of bidirectional frames to find a predictive motion vector set by calculating a temporal distance between a nth reference frame and a reference frame referred by an input motion vector and finding at least one best prediction point motion vector of the nth reference frame from a set of motion vectors by utilizing a cost based motion search. The cost based motion search comprising finding a prediction motion vector, finding a cost factor, performing a cost based motion search of at least one partition level to find the best motion vector from the set of motion vectors.