The theory of “embodied cognition” stresses the roles of the body and the environment in cognitive processing, and prior studies have shown that thoughts, sensations, and behaviors are rooted in interactions between the two. However, it has also been found that several empirical embodiment effects cannot be replicated, or that the replicating effect sizes are far smaller than those of the original effect. Moreover, the significance of an embodiment effect is not always clear. Following a review of the boundary conditions of embodiment effects, the present article proposes a dual process perspective of embodied cognition. The dominating embodied theories incorporate embodied metaphor theory, perceptual symbol systems, embodied simulation theory, enactive cognition theory, and somatic marker theory. They have used conceptualizations such as “mapping,” “simulators,” “sensorimotor simulating,” “perception–action circuits,” and “somatic circuits” as integrating mechanisms with which to link the environment, the body, and cognitive processing. We posit that an embodiment effect is influenced by the relative robustness of these integrating mechanisms. In the early stages of integration, such links are weak; cognitive processing can be influenced through manipulation of the embodied variables, but the effect is unstable and can be easily altered by other stimuli. In the later stages, link strength is enhanced and the effect becomes robust, yet the effect size is reduced. Thus, cognitive processing may be less influenced by embodied variables, but the effect is robust and cannot be easily changed by other stimuli. Alongside variations in embodiment effect size and robustness, dual process types are also developed, and we hypothesize that a dual process could moderate an embodiment effect. Empirically, there is evidence to support this perspective. First, embodiment effects are moderated by age, in that the effect size in adults will be smaller than in children, and smaller in older adults than in younger ones. Second, embodiment effects are moderated by the individual’s processing state, insofar as effects are significant when participants are unconsciously processing the embodied variables, but non-significant when they are consciously processing them. Third, embodiment effects are moderated by whether sufficient cognitive resources are available: when cognitive resources are depleted during an experimental task, the embodied variable cannot be accessible to cognitive processing, and the embodiment effect would be non-significant. Fourth, embodiment effects are moderated by the potential psychological conflict between embodied variables and cognitive processing; if such conflicts are resolved by contextual factors, cognitive processing will not be affected by embodied variables. In our paper, these postulations collectively are advanced as a dual process perspective of embodied cognition, through reference to which we provide several recommendations for future manipulations of embodied experiments, as follows: (a) embodied research requires larger sampling sizes than non-embodied research, since embodiment effect sizes are usually smaller than preceding non-embodiment effect sizes; (b) embodied variables should be well covered to ensure being processed by the intuitive Type Ⅰ process, because, if they were processed by the deliberate Type Ⅱ process, the embodiment effect would be eliminated by cognitive control; (c) the difficulty of the embodied task needs to be properly designed, given that, if it is too difficult, cognitive resources will be depleted and the embodiment effect would be non-significant; (d) psychological conflict between embodied variables and cognitive processing should be retained in the experiment’s procedure, as, thereby, the cognitive processing will be adjusted, following a change of embodied variables, to keep mental processes consistent.