In this paper, we address a restoration problem for graph signals with unknown signal values. First, we introduce an autoregressive model for graph signals. We next develop a restoration method based on a rank minimization technique. The proposed method is demonstrated by applying to a PageRank estimation problem of a web network.