In large-scale multi-objective optimization problems (LSMOPs), the extensive solution space challenges algorithm convergence and requires more resources to maintain solution diversity. Striking a balance between convergence and diversity becomes challenging. This paper presents DPCSA, a dual-population co-sampling algorithm, to tackle this issue. It sustains two populations simultaneously, the primary population and the archive population, which aims to conduct an efficient search and obtain globally optimal solutions. Moreover, to boost the algorithm’s ability to explore, DPCSA introduced a new sampling strategy. This strategy combines the characteristics of the two populations to perform directional sampling in multiple directions, thus broadening the search range and significantly enhancing the algorithm’s capacity for global exploration. A selection strategy based on reference vectors is employed to update the archive population. This approach helps maintain population diversity. The proposed DPCSA is compared with several cutting-edge algorithms on the LSMOP test suite. Experimental outcomes demonstrate that the algorithm is exceptionally competitive in addressing LSMOPs.