Simulations of atmospheric general circulation models' (AGCMs') continental latent heat flux (LH) depend on the land-surface schemes (LSSs), the atmospheric forcing and feedbacks between these. This paper introduces a novel method for evaluating the dependence of 19 Atmospheric Model Intercomparison Project AGCMs' LH on the LSS by excluding the impact of forcing from, and feedback to, the atmosphere. Pseudo LSSs (PLSSs) for LH in the form of multi-variable linear models expressing mean monthly LH as a function of atmospheric forcing are developed. Analysis over three large and climatically diverse river basins shows estimates of mean annual LH from the PLSSs agreeing well with the AGCMs' simulations. Root mean square errors range from 0.4 to 2.2 Wm-2 depending on the region and the AGCM. When the PLSSs are driven by single realisations of atmospheric forcing, different LSSs interact differently with the atmosphere, exhibiting a range of climate sensitivities, and the variability of mean annual LH among AGCMs increases. We conclude that LSSs are important to AGCM predictions and new, or damaged, land-surface components will increase diversity among climate simulations.