It is desirable for AGCMs to evolve atmospheric fields whose statistical properties are similar to obsevations in order to minimize any biases that must be mitigated via assimilation techniques. In this work we examine the three-dimensional distribution of specific humidity in state-of-the-art climate models as compared to measurements from the AIRS satellite system. We find the majority of models have a pattern of drier than observed conditions (by 10-25%) in the tropics below 800 hPa, but 25-100% too moist conditions between 300 and 600 hPa, especially in the extra-tropics. Analysis of the accuracy and sampling biases of the AIRS measurements suggests that these differences are due to systematic model errors. For climate timescale integrations, this error might affect the model-estimated range of climate warming anticipated over the next century.