The plastic multihole products are widely used in Household Appliance and Automobile industry. The densely distributed weld lines resulting from the multiaperture structure seriously damage the surface appearance and mechanical properties of the products. However, there are relatively few studies on the optimization of weld lines of the plastic multihole products. Therefore, a multihole plate was taken as an example to exploit an optimization strategy for this kind of defects. The multi‐objective evaluation system was established to assess bonding quality of weld lines based on Numerical Simulation. And a multi‐objective optimization methodology was presented to efficiently optimize the process parameters. First, back propagation neural networks and Kriging method were combined to build the surrogate models based on the experiments arranged by Latin hypercube sampling. Second, a systematic test strategy was proposed to ensure high accuracy of the surrogate model. Then, the pareto optimal solution was obtained by integrating the surrogate model and Non‐dominated Sorting Genetic Algorithm II. Finally, the simulated and practical confirmation experiments indicate that bonding quality of weld lines can be effectively characterized and significantly improved by the multi‐objective evaluation system and the multi‐objective optimization methodology, respectively. POLYM. ENG. SCI., 59:781–790, 2019. © 2018 Society of Plastics Engineers