To enable a rapid and reliable performance test without obtaining an accurate cross talk value by applying different tensions in the transmission axis direction of an optical fiber, and obtaining plural rough cross talk values corresponding to each tension.

A movable optical-fiber gripper 25 is stopped by a stopper 5 to keep a fiber retention board 1 from sliding, and its clamp member 2 and a clamp member 2 of a fixed optical-fiber gripper 26 are set in the released position. A polarized-wave preservation optical fiber 8 is loaded on a groove 21 and clamped by rotating and closing each clamp member 2. Each polarizer on the incidence and outlet sides are rotated in a state of imparting no tension to the optical fiber 8, and each of the maximum and the minimum analysis quantities are measured to find a rough cross talk value C'1. Thereafter, the stopper 5 is released, and the elastic force of a torsion coil spring 3 is made to function on the fiber retention board 1 to apply a tension on the optical fiber 8, and a rough stroke value C'2 is obtained in the same way. A highly reliable test can be executed by comparing the rough values C'1 and C'2, even when a highly coherent light source is used.