By suitable representation of climate model output, the number of realizations required for meaningful climate model statistics may be significantly reduced, thereby freeing up sorely needed computer time.  It has been noted that model variables when represented spectrally, show amplitude variations significantly more uniform than phase.  We documented this conjecture using archived model simulation data represented in spectral space by the Solid Harmonic functions often used in numerical model integrations.  If it is clearly established that amplitude does not vary substantially amongst the realizations, then in situations wherein amplitude is an adequate measure of climate, many fewer realizations would be needed for defining the model climate variability, thus leading to substantial reduction in computer time usage.  From a many year integration archive of a two level baroclinic model, 80 realizations of the mean and shear stream field were selected as an ensemble data set to test for the relative variability of amplitude and phase. . The results of analyses with this ensemble demonstrated that the amplitudes of variables calculated with the inclusion of phase varied considerably more than amplitudes calculated in isolation, thus corroborating our original hypothesis and indicating that many ensemble calculations are needed primarily to overcome variability in phase predictability.  We demonstrated that these results did not depend significantly on model truncation.  Moreover, the results were confirmed for a state-of the-art model by repeating the analysis with ensemble data available in the NCAR archives from integrations with the CCM3.  The large variability in phase was also documented from independent analyses of spectral phase data.  For climate model results dependent primarily on amplitude, considerably fewer realizations are required to establish model variability.