In this study, we propose an extension of the Penalized Complexity (PC) prior framework to a flexible class of multivariate spatial models for areal data, referred to as M-models. In the spirit of the framework developed by Botella-Rocamora, Martínez-Beneito, and Banerjee (2015), we compare the application of PC priors for the multivariate parameter controlling for the strength of spatial association within three standard models employed in disease mapping literature: the Proper Conditional Autoregressive (PCAR) model, the Leroux Proper Conditional Autoregressive (LCAR) model, and the scaled Besag–York–Mollié (BYM) model. By employing a stepwise procedure, we find that the multivariate PC-prior can be derived analytically and factorised into the univariate marginals of the single parameters, which in turn are independent on any other model parameter.  Presently, we provide an application example on male mortality data in the Valencia region (1987–2011) from three causes of death (cirrhosis, oral cavity cancer, and lung cancer) by using R-INLA.