Freeform surfaces freeform optical surfaces are widely utilized in optical engineering domain, while the traditional computer-aided alignment methods fail to guide the alignment of the optical system containing. A novel method using artificial neural networks is proposed to assist the alignment of optical system. The logical model of alignment with neural networks is introduced, and two alignment simulation examples are taken to verify the practicability of this method. The alignment results show that when the optical path difference distribution or the simulated Zernike polynomial coefficients of the system exit pupil wavefront are used as imaging quality parameters，the root-mean-square errors of misalignment parameters computed by the neural network method are less than 7.04%. The neural network method provides a certain reference to alignment of freeform surface systems.