There have been many formal advances in the analysis of phonological opacity in the recent literature, but experimental studies that shed light on the nature of the generalizations that speakers make about opaque patterns and how opaque forms are accessed in spoken word recognition remain relatively rare. In this paper, I explore how native speakers of various Chinese dialects internalize transparent and opaque tonal alternation patterns (often referred to as tone sandhi) by using wug tests, auditorily primed lexical decision tasks, and acceptability rating tasks for variable tone sandhi forms. In particular, I focus on the comparison between sandhi patterns that can be driven by surface-true phonotactics and those that involve a circular chain shift. One set of comparisons comes from right-dominant tone sandhi patterns in which nonfinal syllables of a tone sandhi domain undergo paradigmatic substitution. Wug test results showed that transparent substitution patterns in Mandarin and Hailu Hakka are generally productive, while substitution patterns that are involved in a circular chain-shift in Taiwanese Southern Min lack full productivity and often categorically fail to apply to nonce words. Auditory priming studies showed that disyllabic words undergoing a transparent sandhi are more strongly primed by a syllable carrying the base tone, while words undergoing a sandhi in a circular chain shift are more strongly primed by the sandhi tone. Another set of comparison comes from left-dominant tone sandhi patterns in which the tone on the initial syllable of a tone sandhi domain is spread over the domain, either directly or after substitution. Wug test results showed that a transparent tone spreading pattern in Shanghai Wu is generally productive, while the substitution aspect of the tone sandhi in Wuxi Wu, which involves a circular chain shift, is not. Variation patterns of tone sandhi in the two dialects also behave differently with respect to how morphosyntactic structure and frequency influence the variation, and the difference can be interpreted as stemming from the higher lexicality of the sandhi forms in Wuxi. These results suggest that at least a subset of opaque alternation patterns calls for an analysis that is morphological/lexical in nature, not a generic phonological solution. The experimental studies, then, provide not only an empirical basis for formal advances of the theory to test their predictions against, but also a guide for the direction of these advances.