This chapter details the parameters that influence the attention model validation. It is structured around four different experiments. Experiment 1 shows that the influence of the ground-truth is not crucial: If models have good results with one ground-truth, it is quite unlikely that these models completely fail with the other ground-truth except due to statistical fluctuation. Experiment 2 shows that the properties of the stimuli (e.g., large, medium, and small salient regions) are addressed with different degrees of accuracy by the classical saliency models (not using deep learning). For eye tracking-based models, small salient regions are better detected than medium and large salient regions. With object detection the exact opposite behavior is observed. However, deep learning models might be much less sensitive to the size of the important objects. Experiment 3 shows that several parameters such as centered bias, saliency map fuzziness, or border cut have an important influence on the final result. Experiment 4 shows that the minimal set of similarity metrics to be used is composed of three carefully chosen metrics which is enough to provide a fair ranking result.