Artificial grammar learning (AGL) is one of the most commonly used paradigms for the study of implicit learning. A well known paradigm to assess learning is artificial grammar learning, originally designed by Reber (1967). In this task, participants are instructed to memorize letter strings. They are not informed that the strings are constructed according to certain rules. In a subsequent grammaticality classification test, the participant’s ability to discriminate between grammatical and nongrammatical items is assessed. Grammaticality judgments can be based on structural (“rule-based”) or superficial (“chunk-based”) aspects of the grammar, or on a combination of both. Artificial grammar paradigm is not only the widely used in implicit learning, but also one of the most important method for the research on the implicit knowledge. We analyze the component materials of the artificial grammar paradigm, for example, letter frequency, chunks, exemplar as well as abstract rules and other factors, and summarize the experiments which various kinds of implicit knowledge is separated in brain neural level in artificial grammar paradigm. We obtain more complete understanding on sequence learning of letter strings. It provides the reference for the further study on the specific problems of knowledge with brain neural mechanism and promotes the further study of unconscious knowledge in artificial grammar paradigm. What knowledge is learned in artificial grammar paradigm by participants depends on the experiment operation of artificial grammar materials. If the factors other than stimulus frequency in the letter strings are balanced, we can discuss the questions whether the participants obtain stimulus frequency knowledge; if the factors other than chunks are balanced, we can discuss the questions whether the participants obtain chunks knowledge; if the factors other than exemplar similarity are balanced, we can study the function of exemplar during implicit learning process. It is the most important for the researcher of implicit learning that they can verify the hypothesis of rule-based acquisition in implicit learning by balancing the factors other than rules. The researchers explore stimulus frequency, chunks etc, on the one hand it prompts us to have a more comprehensive understanding for implicit knowledge, but on the other hand it is beneficial to have a more definitive answer for this hypothesis of rule knowledge.