Listeners use the prosodic cues of an utterance to help determine its syntactic structure, but how does this process happen in the specialized domain of mathematics? Mathematical expressions can contain deeply embedded structures, and listeners encounter read mathematical expressions (RMEs) far less frequently than other potentially ambiguous utterances. How does experience with listening to math affect our ability to hear the structure of an RME via its prosody? Are there limits to the amount of structure we can pull out of the prosody of an utterance? A perception experiment was conducted with subjects aged 7-59 to help answer these questions. Participants heard recordings of RMEs and attempted to determine which of two or more mathematical structures the reader intended. When subjects chose between two options for phrases like nine times A minus two, they chose the mathematical expression that had bracketing matching the prosody of the utterance. However, for more complex phrases like the square root of sixteen over A plus twelve, results were at chance. Age played a surprising role: subjects’ performance increased dramatically from age 7 to 16, but adults’ performance varied widely. This is attributed to variation in exposure to read mathematics.