The current study explores the connection and affordances of talk and gesture in collaborative science problem-solving activities using an emerging analytical approach. A total of 15 three to five-member groups of middle school students participated in a set of clinical collaborative science problem-solving activities. Six groups (three low- and three high-performing groups) were selected for further analyses. The data consisted of successive statements in student conversations that were coded as different types of talk and gestures. Pearson Phi correlation test and Epistemic Network Analysis were used to analyze the data. We found that certain types of gesture were moderately associated with certain types of talk. In addition, we found high-performing groups were significantly different from low-performing groups in the use of higher cognitive engagement and collaborative regulation talks. We did not find a significant difference in the use of gestures on their own. However, we found that when talk and gestures were combined, high-performing groups tended to use higher cognitive engagement and collaborative regulation talks with representing and pointing gestures. These findings support previous studies on the affordances of talk and gesture in collaborative science activities and provide more evidence for how multimodality is important in science learning.