To clarify the source of deviations of drug combination effects evaluated with different drug   interaction mathematical models, the cytotoxicity of SAHA and arsenic trioxide and their combinations were  observed in a series of human cancer cell lines and a normal cell line.  The combined effects were evaluated with three drug interaction models: Loewe Additivity (LA), Bliss Independence (BI) and Chou’s Median Effect Model.  The evaluations with three different models were further compared with each other.  We demonstrated that when dose-response curves were fitted with the same method, similar evaluated results for drug combinations would be derived with different models.  The deviations of evaluated effects of drug combinations were attributed to different curve fitting methods used rather than the models themselves.  The effects of drug combinations showed discrepancies on different cell lines, and at different combined drug concentrations on same cell line.