The precise geometric organization of the cerebellar cortical circuitry has allowed connectivity between the individual neurones to be well studied (Eccles et al. 1966; Ito, 1984). Most electrophysiological analyses of cerebellar processing focus on Purkinje cell activity (Llinas & Sasaki, 1989; Keating & Thach, 1995; Welsh et al. 1995; Lang et al. 1999). While Purkinje cells are the output cells of the cerebellar cortex, these studies usually assume that Purkinje cell activity is a ‘simple read-out’ of the underlying activity in the granular layer. Golgi cells are an important element of the granular layer, being activated by mossy fibres both directly and via parallel fibres (Palay & Chan-Palay, 1974). Golgi cells inhibit local granule cells, so modulate excitatory transmission through mossy fibre–parallel fibre pathways. At the synaptic level, Golgi cell inhibition of granule cells has both fast synaptic and slower extrasynaptic ‘spillover’ components (e.g. Rossi & Hamann, 1998; Mitchell & Silver, 2000; Watanabe & Nakanishi, 2003). The circumstances in which Golgi cell firing is modulated in vivo are poorly understood, therefore their contribution to granule cell and Purkinje cell firing remains elusive. Based on connectivity, and assuming that Golgi cells sample inputs from parallel fibres that also excite local Purkinje cells, it was proposed that Golgi cells mediate feedback inhibition to exert ‘gain control’ over mossy fibre–granule cell transmission. In this scheme, if more granule cells become active, they excite Golgi cells and as a result produce greater inhibition of granule cells, which modulates overall excitation (Marr, 1969; Albus, 1971; Ito, 1984; see Maex & De Schutter, 1998). This feedback inhibition will be accompanied by feedforward inhibition driven by direct mossy fibre connections to the Golgi cells (Precht & Llinas, 1969). Several lines of evidence suggest that Golgi cells are not well suited for gain control, and modelling studies (Maex & De Schutter, 1998; De Schutter et al. 2000) have suggested an alternative role for Golgi cells; that is, to synchronize granule cell activity through closed loops formed by the granule cell–Golgi cell pathway, so that granule cell spiking is precisely timed by Golgi cell inhibition. Such patterns of Golgi cell–granule cell firing should be represented in the simple spike output of Purkinje cells. Modulated activity of putative Golgi cells in vivo has been described during vestibulo-occular reflex adaptation (Miles et al. 1980), locomotion in cats (Edgley & Lidierth, 1987) and limb movement in monkeys (Van Kan & Gibson, 1993). De Schutter and coworkers have investigated the response properties of putative Golgi cells in Crus I and II of the rat cerebellar hemispheres (Vos et al. 1999, 2000; Volny-Luraghi et al. 2002). However, only recently have data been reported from Golgi cells identified definitively using juxtacellular labelling (Simpson et al. 2005). Here we have employed juxtacellular labelling to characterize Golgi cells. We show that in addition to brief excitatory responses from focal, mainly trigeminal afferents (Vos et al. 1999), Golgi cell firing is frequently depressed over several hundred milliseconds after peripheral afferent stimulation, while the same stimuli frequently evoke modest increases in Purkinje cell firing over a similar time course. These responses can be evoked with or without preceding excitations, are of long duration and can be evoked by stimuli from much of the body suggesting a highly convergent sensory input to the cerebellar cortex. These responses have not been described before and imply a functional role for Golgi cells that is very different from current models.