Traditionally, the calyx of Held synapse is viewed as a highly reliable relay in the sound localization circuit of the auditory brainstem, with every presynaptic action potential triggering a postsynaptic action potential in vivo. However, this view is at odds with slice recordings that report large short-term depression (STD). To investigate the reliability and precision of this synapse, we compared slice and in vivo recordings from medial nucleus of the trapezoid body neurons of young adult mice. We show that the extracellularly recorded complex waveform can be used to estimate both presynaptic release and postsynaptic excitability. Whereas under standard slice conditions the synapse underwent large STD, both extracellular and whole-cell recordings indicated that in vivo the size of the EPSPs was independent of recent history. The estimated quantal content was typically <20 in vivo, much lower than in the resting synapse under standard slice conditions. However, due to the large quantal size and summation of EPSPs, the safety factor of this synapse was generally still sufficiently large and postsynaptic failures were observed only infrequently in vivo. When present, failures were typically due to stochastic fluctuations in EPSP size or postsynaptic spike depression. In vivo, the calyx of Held synapse thus functions as a tonic synapse. The price it pays for its low release probability is an increase in jitter and synaptic latency and occasional postsynaptic failures. Copyright