The recent surge of interest in nonazeotropic refrigerant mixtures has generated a need for correlation and prediction of liquid phase information for mixtures. We show that the ideal mixture assumption (the linear weighting of pure liquid properties) can be useful but can also be seriously in error when one of the components is near its critical point, even when the mixture is not near its own critical point. We also suggest that using spline-fits to data in isolation is an ineffective way of using such information. A physical model is suggested for a mixture whose equation of state incorporates all the departures of the mixture from ideality, which allows small, isolated sets of data to be used in predicting other physical properties of both the pure fluids and their mixtures.