I outline how scientifically legitimate psychiatric diagnoses are seen as those which have identified underlying causes. I outline the failure of this approach and the move towards basing psychiatric diagnoses on non-sensitive and non-specific causes. Having outlined a neo-Kantian approach to causes I argue psychiatric diagnoses can be probabilistic bridges between causes and symptoms. Categorical and dimensional diagnoses constitute different causal probabilities. I then show how we can use causes to formulate psychiatric diagnoses but this process needs to be constrained by the aim to limit the number of entities employed. This gives categorical diagnoses a useful role even if the DSM was made dimensional or if RDoC and HiTOP produced relatively homogeneous dimensional diagnoses.