The availability of data from many different sources and fields of science has made it possible to map out an increasing number of networks of contacts and interactions. However, quantifying how reliable these data are remains an open problem. From Biology to Sociology and Economy, the identification of false and missing positives has become a problem that calls for a solution. In this work we extend one of newest, best performing models -due to Guimera and Sales-Pardo in 2009- to directed networks. The new methodology is able to identify missing and spurious directed interactions, which renders it particularly useful to analyze data reliability in systems like trophic webs, gene regulatory networks, communication patterns and social systems. We also show, using real-world networks, how the method can be employed to help searching for new interactions in an efficient way.