In this paper, we propose a variable precision rough set analysis of multiple decision tables. Each decision table is regarded as a collected information about an individual opinion. By a rough set analysis of each decision table, we may capture the core opinion of each individual by the lower approximation and marginal opinion by the boundary region. Based on this idea, we propose a rough set approach to capturing the group consensus or agreeable opinions. To this end, we analyze multiple decision tables. In order to treat the error caused by human evaluation as well as to accommodate disagreements among individuals, we apply a variable precision rough set model. To treat the possible difference of evaluated objects among decision tables, we propose two approaches. In both approaches, we define lower and upper approximations under multiple decision tables and multi-agent rough sets are defined by a pair of lower and upper approximations. Moreover we describe a rule induction method under a multi-agent rough set. The approach is applied to real data about preferences of several students among Japanese companies to be employed.