Listening to music entails the construction of a mental representation based on partial and ambiguous information. This study examines an experimental method that reflects such parsing decisions on-line by detecting the cognitive load resulting from temporary parsing failures. The method investigated was a divided attention paradigm in which listening to music was the primary task and click detection was a concurrent secondary task. It was hypothesized that increasing the complexity of the primary task by introducing an unprepared chromatic modulation results in an increase in response latencies to a click presented immediately after the modulatory shift. The support of this prediction by musicians' data provides evidence for the sensitivity of the paradigm. The failure of non-musicians to reflect the expected load is attributed to their attention-allocation strategy. These results are discussed in terms of their implications on the view of the musical parser as deterministic.