<jats:title>Abstract</jats:title> <jats:p>The accuracy of all types of Vaisala radiosondes and two types of Snow White chilled-mirror hygrosondes was assessed in an intensive in situ comparison with reference hygrometers. Fourteen nighttime reference comparisons were performed to determine a working reference for the radiosonde comparisons. These showed that the night version of the Snow White agreed best with the references [i.e., the NOAA frost-point hygrometer (FPH) and University of Colorado cryogenic frost-point hygrometer (CFH)], but that the daytime version had severe problems with contamination in the humid upper troposphere. Since the RS92 performance was superior to the other radiosondes and to the day version of the Snow White, it was selected to be the working reference. According to the reference comparison, the RS92 has no bias in the mid- and lower troposphere, with deviations &amp;lt;±5% in relative humidity (RH). In the upper troposphere, the RS92 has a ∼5% RH wet bias, which is partly due to the RS92 time lag error and the termination of the heating cycle. It was shown that the time lag effects relating to Vaisala radiosondes can be corrected. Because these were nighttime comparisons, they can be considered to be free from solar radiation effects. Neither the radiosondes nor the Snow White succeeded in reproducing reference class hygrometer profiles in the stratosphere.</jats:p> <jats:p>According to the 29 radiosonde intercomparisons, the RS92 and the modified RS90 (FN) had the best mutual agreement and no bias. The disagreement is largest (&amp;lt;±10% RH) at low temperatures (T ≪ −30°C), where the FN underestimated (overestimated) in high (low) ambient RH. In comparison with the RS92, the RS90 had a semilinearly increasing wet bias with decreasing temperature, where the bias was ∼10% RH at −60°C. The RS80-A suffers from a large temperature-dependent dry bias in high RH conditions, being over 30% RH at −60°C and ∼5% RH near 0°C. The RS80-A dry bias can be almost totally removed with the correction algorithm by Leiterer et al., which was chosen as the best available. The other approach tested tends to overcorrect in high RH conditions when T &amp;lt; −50°C. For T &amp;gt; −30°C it is ineffective and does not correct the RS80-A dry bias in high ambient RH.</jats:p>