In this study, we investigate the performance of GRASP (Greedy Randomized Adaptive Search Procedure) mataheuristic that has been used for solving extended optimization version of CSP (Car Sequencing Problem) with respect to different parameters and proposed modifications for the algorithm. While maximum appearance frequency of option in segment is usually a hard constraint, in this study we treat all capacity constraints as soft constraints by giving them same priority when evaluating candidates for the position in the sequence, in order to achieve more balanced sequences as initial solutions for the local improvement phase. The main disadvantage of the described greedy algorithm using static value of α GRASP parameter is that it uses all the good options at the beginning of the sequence, leaving the worse options for the last part of the sequence. Based on this, changing the starting position of the sequence and order of moves for the local search has been investigated. Results show that the starting position of the sequence for the local search matters and taking into account changing the starting position of the sequence for exploring moves for the local improvement can lead to better solutions.