Many scientific domains use computer simulations as investigative tools (Winsberg 2010). This use presupposes that analysing the results of a simulation can directly or indirectly tell us something about the system represented by that simulation. But what, if anything, validates the inferences drawn from computer simulations? This question can be approached from an applied perspective — that is, how to build an actual simulation that is valid for a specific question under investigation. First, there is the matter of how scientists interact with the simulation outputs: are they treated just like a traditional experiment, are they more akin to a thought experiment or something else entirely? I sustain that understanding simulations as a tool for the construction of thought experiments can explain how simulations can be scientifically relevant even though their design and programming determines their outputs. From that understanding of computer simulations, I then explore how models of data, which Patrick Suppes originally introduced to describe the clash between theories and experimental data (Giere 1999, 54), can help describe the comparison between the outputs of a simulation and the current scientific understanding of its target. By considering those two aspects, we can obtain an image of computer simulations that reflects how those objects are deployed in scientific practice.