A collection of methods for multivariate outlier detection based on a robust Mahalanobis distance is proposed. The procedure consists on different combinations of robust estimates for location and covariance matrix based on shrinkage. The performance of our proposal is illustrated, through the comparison to other techniques from the literature, in a simulation study. The resulting high correct classification rates and low false classification rates in the vast majority of cases, and also the good computational times shows the goodness of our proposal. The performance is also illustrated with a real dataset example and some conclusions are established.