In this paper we propose an asymptotically equivalent single-step alternative to the two-step partially linear model estimator in Robinson (1988). The estimator not only has the potential to decrease computing time dramatically, it shows substantial finite sample gains in Monte Carlo simulations.