This work updates the methods of Lumpkin and Johnson (2013) to obtain an improved near-surface velocity climatology for the global ocean using observations from undrogued and 15-m drogued Global Drifter Program (GDP) drifters. The proposed procedure includes the correction of the slip bias of undrogued drifters, thus recovering about half of the GDP dataset; and a new approach for decomposing Lagrangian data into mean, seasonal and eddy components, which reduces the smoothing of spatial gradients inherent in data binning methods. The sensitivity of the results to method parameters, the method performance relative to other techniques, and the associated estimation errors, are evaluated using statistics calculated for a test dataset consisting of altimeter-derived geostrophic velocities subsampled at the drifter locations, and for the full altimeter-derived geostrophic velocity fields. It is demonstrated that (1) the correction of drifter slip bias produces statistically similar mean velocities for both drogued and undrogued drifter datasets at most latitudes and reduces differences between their variance estimates, (2) the proposed decomposition method produces pseudo-Eulerian mean fields with magnitudes and horizontal scales closer to time-averaged Eulerian observations than other methods, and (3) standard errors calculated for pseudo-Eulerian quantities underestimate the real errors by a factor of almost two. The improved decomposition method and the inclusion of undrogued drifters in the analysis allows resolving details of the time-mean circulation not well defined in the previous version of the climatology, such as the cross-stream structure of western boundary currents, recirculation cells, and zonally-elongated mid-ocean striations. •A new global drifter-based climatology of near-surface currents is presented.•Slip correction reduces differences between drogued/undrogued velocity statistics.•Improved decomposition method reduces the smoothing of spatial gradients.•Statistical errors are found to underestimate real errors by a factor of two.•Mesoscale details are resolved, such as the cross-stream profile of narrow currents.