Spatiotemporal signature is the optic flow specifying the spatiotemporal attributes of a dynamic object that serves for recognition (Stone, 1998). Little is known about the long-term representation of this spatiotemporal signature and its coding nature. The current study investigated the coding of spatiotemporal signature in a dynamic object recognition task. In Experiment 1 and 2, the observers’ recognition performance was impaired by an overall reversal of the studied objects’ motion trajectory as well as the appearance of a novel object, suggesting the spatiotemporal appearance of the objects was coded for recognition. In Experiment 3 to 6, a feature reversal paradigm was applied that only the global-scale or local-scale dynamic feature of the object was reversed at a time. The reversal effect still held but it was selective to the salient feature of the view sequence, suggesting specific spatiotemporal feature of an object instead of its whole view sequence might be used for recognition. Experiment 7 and 8 further generalized the effects of feature-dependent recognition to different learning conditions. Among the 8 experiments, some modulation effects on the coding process from the view sequence (i.e., smoothness, feature’s saliency) was observed and was discussed under the rule of perceptual organization and experience-driven regulation. Taken together, these results supported an emerging framework that argues the spatiotemporal appearance of a dynamic object contributes to its recognition. The framework was extended by the presented evidence that the spatiotemporal signature might be coded in a feature-based manner and under the modulation from both perceptual and top-down regulation.