Agricultural runoff is a leading source of phosphorus (P) pollution to lakes and streams. The objective of this study was to evaluate P removal dynamics in a constructed treatment wetland (CTW) treating agricultural irrigation return flows. The CTW included a sedimentation basin (SB) followed by two surface-flow wetlands in parallel. Typical retention times and total P (TP) loading were 1.4 d and 50 to 110 g m-2 yr-1 P, respectively, for the SB and 5 to 6 d and 4 to 10 g m-2 yr-1 P, respectively, for wetlands. On the basis of this multiyear study, concentration removal efficiency in the SB averaged 21% for TP and 32% for reactive phosphorus (RP). Concentration removal efficiency in wetlands averaged 37 and 43% for TP and 22 and 33% for RP. Areal first-order removal rates for TP averaged 22 and 31 m yr-1 in wetlands. Total P removal in wetlands exhibited a strong seasonal pattern, with minimum removal in the summer when high temperatures likely enhanced P release from decaying plant biomass. The performance of the CTW was stochastic, with removal unpredictably poorer in some years in part as a result of muskrat bioturbation and plant harvesting. In years before muskrat impacts, concentration removal efficiencies in wetlands were 50% for TP and 65% for RP.