For uterine pressure-fetal heart rate data collected during labour and delivery, we first identify the impulse response function (IRF) by least squares. We then use singular value decomposition of this noisy least-squares estimate and an order-reduction search to obtain an improved order estimate compared to that based on the standard minimum-description length (MDL). This ;cleaner' IRF is succinctly parameterized by fitting it with a second-order dynamic model. Using standard hypothesis tests, the model coefficients show statistically significant differences between normal and pathological cases.