have been observed from pulsars aged 10±10 yr, 80 per cent of them come from the youthful pulsars, aged 10±10 yr. For pulsars older than 10 yr, the glitch activity is found to be proportional to the logarithm of the spin-down rate, nC . Based on this relationship, we identify six other pulsars that are most likely to yield frequent large glitches. Since these pulsars are of comparable ages to the Vela pulsar, real-time glitch detection on them and studies of their subsequent recovery would play a vital role in improving understanding of the neutron star interior and the pulsar glitch mechanisms. Key words: methods: statistical ± stars: neutron ± pulsars: general. 1 I N T R O D U C T I O N In addition to a predictable general slow-down in their rotation rate (n ), some pulsars exhibit timing irregularities in the form of glitches or timing noise. A glitch is a spectacular step change in rotation rate, usually accompanied by a change in spin-down rate ( _ n ). While large glitches have a well-defined signature, Dn;D _ n 1;2; microglitches exhibit all possible signatures (Cordes, Downs & Krause-Polstorff 1988). Some of these microglitches have been explained as mere timing noise (e.g. Cordes & Helfand 1980), which is a fairly continuous erratic behaviour in phase, frequency or frequency derivative. In this paper, we have restricted the term glitch to jumps characterized by fractional changes in rotation rate, Dn=n $ 10. It is known that glitches are not due to sudden changes in the external electromagnetic torques on the neutron star, and glitch models are therefore built on changes in the structure of the neutron star or in the distribution and transfer of angular momentum in the neutron star (Alpar 1995). There has been no lack of theories to explain pulsar glitch trigger mechanisms and subsequent postglitch behaviour, as well as timing noise [see D'Alessandro (1996) for a recent review]. Here, our focus is on the `large-glitch' rates of youthful (10± 10 yr) pulsars. In Section 2, we present the glitch magnitudes of all those observed to date. The glitch activity of those pulsars with repeated glitches is used in Section 3 to establish a relationship with observable pulsar parameters. Based on the relationship, we identify pulsars that would make the best targets for frequent observations to detect very large glitches soon after they happen. 2 G L I T C H S I Z E D I S T R I B U T I O N Table 1 shows that 71 glitches have so far been reported in 30 pulsars, representing about 4 per cent of the total pulsar population. About two-thirds of these glitches occur in youthful pulsars, which constitute fewer than half of these 30. The fractional changes in spin-down rate have been measured only in about two-thirds of the reported glitches. The distribution of these jump sizes as a function of age is shown in Fig. 1. Smaller values of Dn /n and D _ n= _ n are seen to be more common in pulsars older than 10 yr. The `Vela-size' jumps Dn=n , 1026 represent about two-fifths of the total glitches in Table 1, the youthful pulsars contributing over 80 per cent of them. None has been reported thus far on very young , 10 yr pulsars. For all the pulsars for which ages have been determined (unpublished catalogue of Taylor et al. 1995), we also find that nearly 70 per cent of the youthful pulsars have glitched (Fig. 2a). The average number of glitches per pulsar is also highest for this age group (Fig. 2b). No glitch has been observed in radio pulsars older than 10 yr and, out of five pulsars younger than 10 yr, only one (the Crab pulsar) has glitched. The surprisingly low level of glitch activity in these very young pulsars has been attributed to their higher internal temperatures which reduce the importance of pinning, resulting in a relatively smooth transfer of angular momentum to their crusts (McKenna & Lyne 1990). Owing to long inter-observation gaps, fractional changes in spin-down rate are poorly measured, with errors in excess of 200 per cent in some cases (e.g. McKenna & Lyne 1990; Shemar & Lyne 1996). This makes DnC/nC poorly suited to a statistical analysis of glitch behaviour. High-precision values of DnC/nC have