Accurately representing aerosol-cloud interactions in global climate models is challenging. As parameterizations evolve, it is important to evaluate their performance with appropriate use of observations. In this investigation we compare aerosols, clouds, and their interactions in three global climate models (GFDL-AM3, NCAR-CAM5, GISS-ModelE2) to MODIS satellite observations. Modeled cloud properties are diagnosed using a MODIS simulator. Cloud droplet number concentrations (N) are computed identically from satellite-simulated and MODIS-observed values of liquid cloud optical depth and droplet effective radius. We find that aerosol optical depth (τ_a) simulated by models is similar to observations in many regions around the globe. For N, AM3 and CAM5 capture the observed spatial pattern of higher values in coastal marine stratocumulus versus remote ocean regions, though modeled values in general are higher than observed. Aerosol-cloud interactions were computed as the sensitivity of ln(N) to ln(τ_a) for coastal marine liquid clouds near South Africa (SAF) and Southeast Asia (SEA) where τa varies in time. AM3 and CAM5 are more sensitive than observations, while the sensitivity for ModelE2 is statistically insignificant. This widely used sensitivity could be subject to misinterpretation due to the confounding influence of meteorology on both aerosols and clouds. A simple framework for assessing the sensitivity of ln(N) to ln(τ_a) at constant meteorology illustrates that observed sensitivity can change from positive to statistically insignificant when including the confounding influence of relative humidity. Satellite-simulated versus standard model values of N from CAM5 are compared in SAF; standard model values are significantly lower with a bias of 83 cm^-3.