1. The reflex responses to stretch were studied in single motor units and the surface electromyogram in human masseter. 2. Controlled stretches of the isometrically contracting jaw-closing muscles evoked short-latency (10-15 ms) and long-latency (35-70 ms) excitatory reflex responses in the masseter surface electromyogram. 3. The majority (65%) of tonically active masseter motor units were excited in both short- and long-latency phases of the reflex. The timing of the stimulus determined whether the unit discharged in the short- or long-latency phase. If a non-tonically active motor unit was recruited by the stimulus, it invariably discharged in the long-latency phase. 4. Although short-latency responses were strongly time-locked to the stimulus, there was very little shortening of interspike intervals (ISIs) in this phase of the reflex. The shortening of ISIs was more prominent and prolonged during the long-latency phase, which explains why this phase produces most of the reflex force changes following the stretch. 5. Within pairs of concurrently active motor units there was a tenfold range in the size of the short-latency response to the same stretch. 6. A substantial proportion (35%) of the twenty-two masseter motor units tested had no statistically significant short-latency reflex response. 7. In contrast to other human muscles, there was no functional connection between a population of Ia afferents and some masseter motoneurons. There are two possible explanations for this result. The short-latency, presumably monosynaptic, Ia afferent inputs may not be uniformly distributed to human masseter motoneurons. Alternatively, these inputs may be subject to tonic presynaptic inhibition that is not uniformly distributed throughout the masseter motoneuron pool.