Shape puzzles can elicit parent-child math talk, which is critical for early math learning. However, little is known abouthow the features of the puzzles impact parent-child interactions through parents math talk. Two- to four-year-old chil-dren and their parents (current N=30; target N=128) completed two shape puzzles. The control puzzle was typical ofcommercial puzzles, including nine distinct shapes. The experimental puzzle included multiple exemplars of shapes (e.g.,three different triangles, three different quadrilaterals). We hypothesized that parents would use richer math talk with theexperimental puzzle. We coded quantity and quality of parent math talk during the interactions. Preliminary results indi-cate that parents mostly used low-level math talk (naming shapes) for both puzzles, but they used more high-level mathtalk (comparing shapes, providing shape definitions) for the experimental than the control puzzle (p=0.054). We discussparticular puzzle features that can stimulate high-quality math talk during parent-child interactions.