The need for accurate computational procedures to evaluate detailed properties of gas phase chemical reactions is evident when one considers the wealth of information provided by laser, molecular beam and fast How experiments. By stressing ordinary scalar computers to their limiting performance quantum chemistry codes can already provide sufficiently accurate estimates of the stability of several small molecules and of the reactivity of a few elementary processes. However, the accurate characterization of a reactive process, even for small systems, is so demanding in terms of computer resources to make the use of supercomputers having vector and parallel features unavoidable. Sometimes to take full advantage from these features all is needed is a restructure of those parts of the computer code which perform vector and matrix manipulations and a parallel execution of its independent tasks. More often, a deeper restructure has to be carried out. This may involve the problem of choosing a suitable computational strategy or the more radical alternative of changing the theoretical treatment. There are cases, in fact, where theoretical approaches found to be inefficient on a scalar computer exhibit their full computational strength on a supercomputer.