This study examined how Chinese children acquire the untaught positional constraints of stroke patterns that are embedded in left-right structured and top-bottom structured characters. Using an orthographic regularity pattern elicitation paradigm, 536 Hong Kong Chinese children at different levels of reading (kindergarten, 2nd, and 5th grades) were asked to produce invented characters with left-right and top-bottom stroke pattern pairs. Even kindergartners were aware of the positional constraints of stroke patterns and were able to produce orthographically legal pseudocharacters with different stroke pattern pairs. This ability improved across grade level. Moreover, there was a production asymmetry in which children produced more top-bottom structured pseudocharacters than left-right structured pseudocharacters. The error pattern analysis further revealed that more positional errors were observed in producing left-right structured noncharacters than in the top-bottom structured noncharacters. This production asymmetry seemed to reflect children's experience with a distribution asymmetry observed between left-right (59.19%) and top-bottom structured characters (23.46%) in a corpus of school Chinese. These results are discussed within the framework of statistical learning of orthographic regularity in Chinese.