Based on a review of empirical studies in over thirty years, this paper sketches three developmental stages of complex words processing theories. The first stage is characterized by a debate on whether complex words are accessed as a whole or as more than one unit, i.e., whether component morphemes are involved in the processing or not. Major conflict is reflected in the debate between the decomposed account (Prefix-Stripping Model, Taft & Forster, 1976; Morphologically Decomposed Account, Zhang & Peng, 1992) which assumes that complex words are represented as two units in the mental lexicon and thus are accessed as two components, and the Full Listing Account which contends that complex words are represented and thus accessed as unitary units, and that morphemes do not play a role in complex words processing. As more and more studies (Ford et al., 2010; Yen et al., 2008) indicate that morphemes are involved in complex words processing and therefore rule out the Full Listing Account, the second stage switches to the nature of morphological processing during the complex words access—whether morphological processing is based on form or meaning. The debating parties are known as morpho-orthographic decomposition account and morpho-semantic activation account, both focusing on semantic transparency effect of morphological processing yet reaching opposite conclusions. Specifically, the former argues that, morphological processing is based on form because transparent and opaque complex words produce similar priming effect to their root; while the latter contends that, morphological processing is based on semantics because transparent complex words produce significantly larger priming effect to their root than opaque words. However, studies supporting the morpho-orthographic decomposition account tend to favor long prime duration as adopted in the cross-modal priming and the unmasked priming; while those for morpho-semantic activation account usually take short prime duration as adopted in the masked priming. This seems to suggest that at very early stage of complex words processing, morphological processing is mainly orthographic decomposition, while the meanings of component morphemes are not activated until a later stage. Such observations are consistent with a form-then-meaning account to which evidence has been given (Longtin et al., 2003). Major challenge against the form-then-meaning account comes from the observation of the morpheme meaning activation at very early stage of complex words processing (Feldman et al., 2009), which embraces a form-with-meaning account proposed in the Hybrid Model (Diependaele et al., 2005; 2009). The main dispute of the third stage is between form-then-meaning account and form-with-meaning account. Finally the paper discusses limitations in previous studies and directions for future study. It indicates that in order to solve the dispute between form-then-meaning account and form-with-meaning account, future researches must examine whether morpheme meanings are activated during complex words access and if yes, its time course.