Probabilistic programming libraries like PyMC3, PyStan, and Edward provide flexible ways for users to define Bayesian models, and powerful algorithms for generating samples from the posterior distribution of these models in the presence of observed data. These posterior samples are naturally high dimensional, and each library has a different strategy for handling and inspecting this data. ArviZ is an open source collaboration between PyMC3 and PyStan developers that uses xarray as a common data structure. ArviZ implements common visualization and criticism tasks, useful in making sense of these models.