Hays (1995) and Hays et al. (1996) analyzed Continuous Plankton Recorder (CPR) records to determine if die1 vertical migrations (DVMs) of North Sea copepods had changed in response to changes in predator (chiefly herring) abundances between 1958 and 1996. While they did find such a correlation, their analysis has been challenged by Hirst & Batten (1998), primarily on 2 grounds: (1) Hays (1995) and Hays et al. (1996) did not properly assess the dark period, mistakenly including some light periods as dark and vice-versa, due to seasonally changing day-lengths; (2) they used an index of diel migration, their 'DVMindeX1 (changed in the second paper to 'N/D,nde,'), which was sensitive to sample size. Hirst & Batten (1998) proposed their own, which they confusingly termed 'DVM' (the usual abbreviation for the phenomenon of diel vertical migration itself-] will refer to the index as 'HB'), designed to remedy the sample-size sensitivity in N/Dmde,, the Hays index. In fact, as a ratio measure, N/D,,d,, should not be sensitive to abundances, but only becomes so because of the '1' added to allow logarithmic transformation when abundances are zero. In general, CPR sample sizes are too large for any biases to be noticeable. The variability at very small sample sizes could be reduced by use of a constant small relative to the counting unit, say 0.001, instead of 1. If this were done in the example calculation provided by Hirst & Batten, then when n,,,,, = 20 and nd,, = 10, an N/Dhd,, of 1.9999 would be estimated (instead of 1.909), while with nnlghl = 2 and nd,, = 1, an N/D,,d,, of 1.9990 would be estimated (instead of 1.5). Hirst & Batten (1998) go on to test a different hypothesis: that it is the availability of food (as water color) which governs DVM behavior (Hardy & Gunther 1935). Though not quite significant, correlation of their index HB with water color was indeed better than with herring abundance, as would be expected from results from other studies (e.g. Huntley & Brooks 1982, Johnsen & Jakobsen 1987, Flik & Ringelberg 1993). However, it remains reasonable that effects on migration of predator abundance should be noticeable at least in a long time-series (e.g. see Gliwicz 1986). It would be interesting to investigate the combined effect of both factors. However, neither N/D,nde, (a ratio) nor HB (a proportion) can be considered normally distributed, so that, strictly, neither should be used in ordinary parametric statistical analyses. Thus, probabilities based on associations of either index with environmental variables, while useful guides, can only be considered approximate. A better approximation could be achieved by using a 'normalizing' transformation, such as the arcsine, on HB (Sokal & Rohlf 1981), or by using a nonparametric test. Other, potentially more serious problems come from the sampling methodology itself. First, variations in abundance at a single, near-surface depth could easily be the result of patchiness or other artifacts (Roe 1974). Only strong repeatability (consistently finding higher abundances at night) renders this a plausible method of DVM investigation. As the number of days covered by a sampling program is decreased, eventually there would be no way to identify the signals as true DVM. The depth sampled is usually about 6.5 m (Hays & Warner 1993), not really a 'surface' sample, and it is seldom clear whether the copepods were feeding at this depth, merely passing through, or engaging in 'reverse' DVMs from the surface layer (Roe 1974, Hays et al. 1996). Even assuming that the copepods do spend the dark hours within the phytoplankton maximum, as the depth of this varies through the year, the perceived migration regime must vary with it. Second, and more importantly, it is also known that many zooplankton species undertake other patterns of diel excursions. A common pattern, known at least from the time of Michael (1911), involves 'midnight sinking', in which the population comes to the surface near dusk, then descends again and may reappear at the surface later ('dawn rise') before making its daytime descent. This bimodal activity cycle actually