We compare the combined distribution of 31 group and 25 cluster velocity dispersions [Zabludoff et al., AJ, 106, 1301 (1993)] with the ensemble of 32 models for the formation and evolution of large-scale structure examined by Weinberg & Cole [MNRAS, 259, 652 (1992)]. The models include Gaussian and non-Gaussian initial fluctuations, different power law spectra (n=-1, n=0, n=-2, ''pancake''), flat (OMEGA=1) and open (OMEGA=0.2) cosmologies, and unbiased (b8= 1) and biased (b8=2) galaxy formation. The set of initial conditions we test, although limited, samples enough parameter space to indicate which general classes of models are consistent with the data. The two Gaussian, n=-1 models which best approximate the standard and open Cold Dark Matter (CDM) models do not match the observed distribution of velocity dispersions; models with b8=2 and OMEGA=1 (''standard'') or b8=1 and OMEGA=0.2 (''open'') predict too large a ratio of low to high velocity dispersion systems. A ''COBE-normalized'' CDM model with b8=1 and OMEGA=1 produces clusters with velocity dispersions higher than those measured. All three models overestimate the total abundance of systems.