This work demonstrates a fast, sensitive method of characterizing the dynamic performance of MR gradient systems. The accuracy of gradient time-courses is often compromised by field imperfections of various causes, including eddy currents and mechanical oscillations. Characterizing these perturbations is instrumental for corrections by pre-emphasis or post hoc signal processing. Herein, a gradient chain is treated as a linear time-invariant system, whose impulse response function is determined by measuring field responses to known gradient inputs. Triangular inputs are used to probe the system and response measurements are performed with a dynamic field camera consisting of NMR probes. In experiments on a whole-body MR system, it is shown that the proposed method yields impulse response functions of high temporal and spectral resolution. Besides basic properties such as bandwidth and delay, it also captures subtle features such as mechanically induced field oscillations. For validation, measured response functions were used to predict gradient field evolutions, which was achieved with an error below 0.2%. The field camera used records responses of various spatial orders simultaneously, rendering the method suitable also for studying cross-responses and dynamic shim systems. It thus holds promise for a range of applications, including pre-emphasis optimization, quality assurance, and image reconstruction.