Since the concept of a rotational thunderstorm was presented by Byers in 1942, little attention has been paid to this important characteristic. Through direct and indirect observations, as well as a series of numerical experiments, the authors, some 24 years later, now postulate that many large thunderstorms are rotating. The numerical experiments revealed that a thunderstorm in a strong environmental wind field deviates to the left of the mean wind unless it rotates slowly and cyclonically. It was also found that the maximum deviation, either to the right or left, occurs when such a thunderstorm rotates with a critical tangential speed of only a few meters per second. This striking result contradicts the conventional expectation that the faster the rotation, the larger the storm's deviation. Further investigation of numerically produced clouds revealed that most of the peculiar motion of thunderstorms can be simulated by computing the momentum of clouds through step-by-step integration. A thunderstorm couplet formed by an echo split was successfully simulated numerically. Then the tracks of both cyclonic and anticyclonic storms, almost identical to those observed by radar, were obtained by a computer. The experimental results in comparison with actual storms lead us to conclude that a cloud cannot be treated as a well-mixed entity and that it does not deviate accidentally. Its motion is a consequence of various parameters, including slow rotation, mostly cyclonic but occasionally anticyclonic.