A robot system comprising a first robot (R1) with a camera of a visual sensor mounted thereon and a second robot (R2) having a feature portion, is disclosed. The robots (R1, R2) are set in the first initial states (G1), from which the first robot (R1) or the second robot (R2) is moved so that the image of the feature portion assumes a target position or size (G2), thereby to store the present positions (P1, Q1) (G3). The same process is repeated N times (N.gtoreq.3) while changing the positions of the initial states of the robots (G4). Based on the positions P1, . . . , PN and Q1, . . . , QN obtained by the N repetitive above processes, a matrix T indicating the coordinate conversion from .SIGMA.b to .SIGMA.b' is determined. One or two robot control units may be provided. As a result, the calibration to determine relative positions between the robots can be carried out easily and with high accuracy, thereby reducing the jig-related cost.