Case-Based Forecasting is a useful method for predicting the future using similar cases. The predictive power of a case-based forecasting system depends on (1) accurate similarity measurement between cases, (2) the number of similar cases to combine, and (3) the combining method that assigns weights to similar cases. Among these factors, the number of similar cases to combine has the greatest impact on predictive power. This paper first discusses the three factors affecting the predictive power of case-based forecasting systems, and then examines in greater depth the most important factor—determining the number of cases to combine. Specifically, several methods that can be used to determine the number of cases to combine were compared: (1) the Fixed Number Combining Method, which combines an arbitrary number of cases; (2) Optimal Spanning Methods, which search for the optimal range; and (3) a Mathematical Programming Model using Similarity Distribution. Their effectiveness was compared and analyzed using simulation data. The simulation results showed that the Mathematical Programming Model based on the similarity distribution of cases performed best in most scenarios.