The invention relates to a PET image reconstruction method and system. The method comprises the steps that 101, PET computing equipment acquires detection data in detection equipment; 102, the PET computing equipment acquires a regularization target function phi(x, y) = L(x, y)-beta * R(x) for reconstructing an image according to the detection data, wherein L(x, y) is a likelihood function item corresponding to the detection data, R(x) is a priori knowledge item, and beta is a regularization parameter and is used for adjusting the weights of L(x, y) and R(x); 103, the PET computing equipment determines a parameter beta in a regularization target function according to the signal-to-noise ratio information and the input parameter of the detection data; 104, the PET computing equipment reconstructs a PET image according to the determined parameter beta and the regularization target function, and finally the purpose of improving the quality of the reconstructed PET image is achieved.