Three types of diel vertical migration behavior were expressed within a fjord population of the calanoid copepod Pseudocalanus newmani: (1) reverse migration (nocturnal descent), (2) normal migration (nocturnal ascent), and (3) no detectable migration. Reverse migration by P. newmani occurred in all three study years, but only at a deep (185 m) station at times of year when normally migrating predatory Zooplankton (the copepod Euchaeta elongata, the chaetognath Sagitta elegans, and the omnivorous euphausiid Euphausia pacified) were abundant. A predator-exclusion "natural experiment" tested the hypothesis that the reverse migration is a predator avoidance mechanism: at a shallow (55 m) station within the same fjord, large predatory Zooplankton were absent or reduced markedly in abundance. In conditions of reduced nocturnal predation by predatory zooplankton, no reverse migration by P. newmani was detected on any of 11 cruises, consistent with the hypothesis. Instead, at the shallow station, where diurnal predation by planktivorous fish was of increased importance, normal diel vertical migration by P. newmani was observed on several occasions. No migration was detectable at either the deep station or the shallow station at times of year of low feeding activity of predators in surface waters. The predominant migration behavior (reverse, normal, or no migration) varied depending whether the primary source of mortality was predation by nocturnally feeding Zooplankton or visually hunting planktivorous fish, and appeared unrelated to resource distributions or temperature gradients. The relative advantage of the three migration behaviors is explored with theoretical life table analyses. Re-examination of the demographic advantage hypothesis proposed by McLaren (1974) illustrates that vertical migration across a thermocline carries a fitness cost, not a benefit. The realized rate of increase of a migrant population of Pseudocalanus is found to be lower than a nonmigrant population for all combinations of surface temperatures, positive thermal stratification, and ratios of mortality for early and late developmental stages. However, avoidance of predators by vertical migrants reverses the outcome of this analysis. Predator avoidance by Pseudocalanus, accompanied by reduced mortality rates, confers a fitness advantage upon a population that migrates across a thermocline. Remarkably small reductions in mortality rates (as little as 12%) confer this advantage, whether migrants avoid nonvisually hunting predatory Zooplankton by night or visually hunting planktivorous fish by day. The behavioral variation documented here confirms that diel vertical migration is dynamic rather than a fixed, invariant behavioral trait within a population. In environments where predation pressure and other factors vary through time, studies of limited scope and short duration are unlikely to illuminate the ultimate causes for diel vertical migration. The heritability of the vertical migration trait is unclear. The different phenotypic responses reflected within this single population of P. newmani may reflect altered behavior of individuals. Alternatively, this behavioral variation may represent a balanced polymorphism maintained by differential mortality of genotypes experiencing selection pressures that vary through time.