A multitude of models have been proposed to account for the neural mechanism of value integration and decision making in speeded decision tasks. While most of these models account for existing data, they largely disagree on a fundamental characteristic of the choice mechanism: independent versus different types of competitive processing. Five models, an independent race model, 2 types of input competition models (normalized race and feed-forward inhibition [FFI]) and 2 types of response competition models (max-minus-next [MMN] diffusion and leaky competing accumulators [LCA]) were compared in 3 combined computational and experimental studies. In each study, difficulty was manipulated in a way that produced qualitatively distinct predictions from the different classes of models. When parameters were constrained by the experimental conditions to avoid mimicking, simulations demonstrated that independent models predict speedups in response time with increased difficulty, while response competition models predict the opposite. Predictions of input-competition models vary between specific models and experimental conditions. Taken together, the combined computational and empirical findings provide support for the notion that decisional processes are intrinsically competitive and that this competition is likely to kick in at a late (response), rather than early (input), processing stage. (Contains 22 figures, 5 tables and 16 footnotes.)