Sea-surface salinity (SSS) observed by Aquarius was compared with global observations from Argo floats and offshore moored buoys to evaluate the quality of satellite SSS data and to assess error structures. Aquarius products retrieved by different algorithms (Aquarius Official Release version 3.0 [V3.0], Combined Active-Passive [CAP] algorithm version 3.0, and Remote Sensing Systems test bed algorithm version 3) were compared. The Aquarius SSS was in good agreement with in situ salinity measurements for all three products. Root-mean-square (rms) differences of the salinity residual, with respect to Argo salinity, ranged from 0.41 to 0.52 psu. These three Aquarius products exhibit high SSS deviation from Argo salinity under lower sea-surface temperature conditions (10 m s(-1)), probably due to model bias and uncertainty associated with sea-surface roughness. Furthermore, significant SSS differences between ascending (south-to-north) and descending (north-to-south) paths were detected. The monthly averaged Aquarius SSS (1 degrees x 1 degrees grid) was also compared with outputs from the ocean data optimal interpolation (OI) system operated by the Japan Agency for Marine-Earth Science Technology (JAMSTEC) and the ocean data assimilation system used by the Meteorological Research Institute, Japan Meteorological Agency (MRI/JMA). Negative bias, attributed to near-surface salinity stratification by precipitation, was detected in tropical regions. For 40 degrees S-40 degrees N, rms difference, with respect to JAMSTEC OI, is 0.27 psu for the V3.0, while the CAP product rms difference is only 0.22 psu, which is close to the Aquarius mission goal. Key Points Sea-surface salinity (SSS) observed by Aquarius was evaluated using in situ data SSS retrieved by three different algorithms were compared with each other Error structure of the retrieved SSS was characterized by residual analyses