This paper makes three contributions. First, experiments have shown that simulation procedures used in existing localization algorithms for mobile sensor networks cannot output stable statistical data. This paper discusses the reasons for this and proposes a quantitative method to set up a simulation procedure that can output stable statistical data. Then, the paper evaluates and compares the accuracy of typical localization algorithms for mobile sensor networks in both obstacle-free and non-free environments. Results show that in environments with obstacles, many techniques that have been proposed, in the past, to improve localization accuracy in existing algorithms are useless and inversely decrease the algorithm’s accuracy. At last, this paper proposes several metrics that can be used by a single node to evaluate the accuracy of its location estimate. Results show that the “possible maximum localization error” metric, which was proposed in previous works, performs best by indicating the accuracy of location estimate for a single node.