Snow insulates and changes ice albedo, therefore the precipitation phase identification scheme is important when modeling lake and sea ice growth. Precipitation phase separation schemes in coupled atmospheric-ice models are usually based on air temperatures but, snow fractions as a function of air temperature vary between models. Two examples of models which use 2-temperature thresholds, one for all rain and one for all snow with a linear decrease in snow fraction in-between, are the CAM-3 model used by National Centre for Atmospheric Research NCAR and the coupled Ocean Sea-Ice Model for Earth Simulators (OIFES). CAM-3 simulates 50% snow at 0°C while OIFES simulates 0% snow at the same temperature. Forty-five years of three-hour man-made precipitation phase observations for nineteen Swedish meteorological stations were used to compare different phase separation schemes. Observations of mixed precipitation were included (assumed to be half rain and half snow). A larger fraction (about 70%) of the precipitation was found to be snow at zero degrees as compared to the fractions simulated with the models mentioned above. This indicates that too large a fraction of the precipitation is classified as rain in these models. Consequently they underestimate the insulation of the snow as well as the albedo. For example, the reduction in (conduction driven) ice growth for a 0.5-m ice with 0.1-m low density (100 kg m-3) snow cover is about 90% compared to pure ice. Solar radiation absorption on the other hand is overestimated and this counterbalance might explain why the models perform fairly well with regard to ice growth even if the snow fraction is underestimated.