The choice of outbreak detection algorithm and its configuration result in variations in the performance of public health surveillance systems. The ability of predicting the performance of detection algorithms under different circumstances will guide the method selection and algorithm configuration. Our work characterizes the dependence of the detection performance on the type and severity of outbreak. We examined the influence of determinants on the performance of C-algorithms and W-algorithms. We used Bayesian Networks to model relationships between determinants and the performance. The results on a sophisticated simulated data set show that algorithm performance can be predicted well using this methodology.