The LBL infiltration model is a simplified method for combining weather information with air tightness to calculate residential air infiltration. In this report, we compare infiltration-model predictions with data collected by our Mobile Infiltration Test Unit (MITU), a full-scale test structure that gathers detailed weather and infiltration measurements. To probe for sources of prediction errors, we examine four simplifying assumptions made in the LBL infiltration model : 1) that the flow through the cracks in the building shell can be approximated by orifice flow, 2) that wind-induced and stack induced infiltration can be added in quadrature, 3) that wind-induced infiltration can be represented by averaging the values for three aspect and typical ratios, 4) that wind-induced infiltration can be represented by averaging the values for all wind directions. We use comparisons with measured data to examine these effects qualitatively, and use detailed computer simulations of infiltration in MITU to quantify each effect. The effects of each assumption are represented by the bias and scatter. (The bias is the average error, and the scatter is a measure of the ability of a model to track short-term fluctuations in infiltration rate.) We show that the orifice flow assumption causes an 8% bias and a scatter of 20%, and that the quadrature assumption causes consistent overprediction (bias = 12%, scatter = 6%). For MITU, averaging over aspect ratio causes some overprediction (bias = 11%, scatter = 14%). Although it reduces the ability of the model to track infiltration (scatter = 19 averaging over wind direction has little effect on the mean infiltration rate (bias = 0). When compared to measured data, the LBL model has a bias of 10% and scatter of 28%.