Despite the increasing number of studies on nitrogen (N) and phosphorus (P) removal in free water surface (FWS) wetland systems, there is still a gap in understanding the influence of design variables on the system performance. To address this, we conducted a global meta-analysis of 73 studies employing advanced statistical techniques, kinetic models, and machine learning along with variable importance analysis. The results indicated that random forest (R 2 = 0.55−0.77) and artificial neural network (R 2 = 0.5−0.85) were the best fitting models for TN and TP removal in pilot-scale and large-scale systems. Moreover, permutation importance results using different wetland design variables indicated that the inflow concentration, plant coverage, hydraulic loading rate, and system area are considered the most important variables for N and P removal under large-scale conditions, while the hydraulic retention time, inflow concentration, and water depth are deemed the most important variables under pilot-scale conditions. Furthermore, the removal of N and P was higher in pilot-scale (54.6% and 56.7%) systems compared to that in large-scale (29.0% and 41.9%) systems. Also, the interactions between design variables and the removal process of N and P were investigated to better understand the specific roles of these variables in improving the removal performance.