This report describes the theory for calibration of radar cross section (RCS) in arbitrary monostatic and bistatic geometries. The ambition is not only to find a method where one set of data can be used for the calibration of bistatic data collected in various geometries, but also to find a method where the contributions from the calibration data to the total errors of the calibrated data are as low as possible. It is therefore a prime concern to find a calibration data set with minimum contributions from background, multipath propagation and crosstalk. When calibration objects are used these may also contribute additional background from the target support, errors in the orientation of the object and errors in the assumed object RCS. In order to reduce these errors a method is therefore proposed where calibration data is gathered with the antennas separated and directed towards each other. In order to monitor the stability of the system and to correct for drift reference measurements through a shunt path are introduced. The method also provides ample opportunities to adjust the calibration signal level with respect to saturation and noise levels and to simplify the procedures.