Cloud Computing has emerged as an interesting alternative for running business applications, but this might not be true for scientific applications. A comparison between HPC systems and cloud infrastructure not always sees the latter winning over the former, especially when only performance and economical aspects are taken into account. But if other factors, such as turnaround time and user preference, come into play, the landscape of the usage convenience changes. Choosing the right infrastructure, then, can be essentially seen as a multi-attribute decision-making problem. In this paper we introduce an evaluation model, based on a weighted geometric aggregation function, that takes into account a set of parameters, among which job geometry, cost, execution and turnaround time. The notion of user preference modulates the model, and allows to determine which platform, cloud or HPC, might be the best one. The model has then been used to evaluate the best architecture for several runs of two applications, based on two different communication models. Results show that the model is robust and there is a not negligible number of runs for which a cloud infrastructure seems to be the best place for running scientific jobs.