In the traditional methods of program evaluation and review technique (PERT) network optimization and compression of time limit for project, the uncertainty of free time difference and total time difference were not considered as well as its time risk. The anthors of this paper use the theory of dependent-chance programming to establish a new model about compression of time for project and multi-objective network optimization, which can overcome the shortages of traditional methods and realize the optimization of PERT network directly. By calculating an example with genetic algorithms, the following conclusions are drawn: (1) compression of time is restricted by cost ratio and completion probability of project; (2) activities with maximal standard difference of duration and minimal cost will be compressed in order of precedence; (3) there is no optimal solutions but noninferior solutions between chance and cost, and the most optimal node time depends on decision-makers preference.