Having assessed the performance gains due to evidence fusion, previous works reported contradictory conclusions. For some, a consistent improvement is achieved, while others state that the fusion of a stronger and a weaker biometric expert tends to produce worst results than if the best expert was used individually. The main contribution of this paper is to assess when improvements in performance are actually achieved, regarding the individual performance of each expert. Starting from readily satisfied assumptions about the score distributions generated by a biometric system, we predict the performance of each of the individual experts and of the fused system. Then, we conclude about the performance gains in fusing evidence from multiple sources. Also, we parameterize an empirically obtained relationship between the individual performance of the fused experts that contributes to decide whether evidence fusion techniques are advantageous or not.