With point source pollution gradually being controlled, non-point source pollution has become a threat for water quality in most regions of China. Best management practices (BMPs) have been regarded as the most effective way to control non-point source pollution. However, the effectiveness of regional cropping systems, cultivation methods, policies, and economic costs lead to difficulties in BMP allocation at the watershed scale. In particular, the difficulties further increase with the changes at spatial scale. As a result, the allocation of BMPs has been moved to a multi-objective decision optimization program, to achieve water quality improvement targets under limited inputs. Therefore, there is a need for multi-objective collaborative optimization of BMPs at different spatial scales. Herein, we reviewed the current research pertaining to multi-objective collaborative optimization of BMPs for non-point source pollution with respect to three aspects: identify the critical source areas (CSAs) of non-point source pollution, assessment of BMPs cutting efficiency, and imitate multi-objective collaborative optimization of BMPs. The results indicate that: i) building multi-scale model coupling system, including land scale and watershed scale, would be the most efficient way to accurately identify CSAs; ii) reducing time lag, uncertainty, and spatial and temporal heterogeneity as well as the risk of pollution to improve water quality will be the key to the cutting efficiency of BMPs; iii) building a nonlinear response relation between watershed pollutant reduction and water quality improvement is essential. The BMP database, cost database, and scientific allocation schemes based on evolutionary algorithm (EA) can be combined to build a multi-scale decision supporting system. The allocation scheme of BMPs and the optimum curve of multi-scale cost-effectiveness can then be acquired.