There are two different assertions on the rolling error in the solid-controlled gyro compass which contains two rotors in its inner gyro sphere. One assertion is that there is a rolling error and the other is that there is no rolling error. This paper examines the rolling error caused by the centrifugal force by the experiment to reveal that the first assertionis reasonable, and it also attempts to explain qualitatively how the rolling error occurs. The Hokushin-Plath gyro compass is chosen as a model. The rolling error is examined by the swing test in various periods. From the tests, the following results are obtained. As long as the swing is continued under the fixed condition of the ship's heading, the swinging period and the amplitude, no error appears. In case the gyro compass is affected by the swingings except those of the cardinal planes, the error starts to appear only after the swing is finished and it is increasing slowly. It takes about 20 minutes for the error to reach its maximum value. The type of this error is a quadrantal one which makes the ship's heading high in the first and third quarters and low in the second and fourth quarters. But in each case the experimental maximum error is greater than the theorectical one.