Many populations have consistently biased adult sex ratios with important demographic and evolutionary consequences. However, geographical variation, the mechanisms, temporal dynamics and predictors of biased sex ratios are notoriously difficult to explain. We studied 334 wild populations of four species of African annual fish (Nothobranchius furzeri, N. kadleci, N. orthonotus, N. rachovii) across their ranges to compare their adult sex ratio, its seasonal dynamics, interpopulation variation and environmental predictors. Nothobranchius populations comprise a single age cohort and inhabit discrete isolated pools, with wide-ranging environmental conditions (habitat size, water turbidity, structural complexity, predators), making them ideal to study adult sex ratio variation. In captivity adult sex ratios were equal. In natural populations, adult sex ratios were biased 1:2 toward females in three study species while N. kadleci had sex ratios at unity. There was a decline in the proportion of males with age in one species, but not in the other species, implying most severe male mortality early after maturation, declining later perhaps with a decrease in male abundance. In general, the populations at vegetated sites had relatively more males than populations at sites with turbid water and little vegetation. Selective avian predation on brightly coloured male fish likely contributed to female dominance and vegetation cover may have protected males from birds. In addition, an aquatic predator, a large belastomid hemipteran, decreased the proportion of males in populations, possibly due to greater male activity rather than conspicuous colouration. Alternative explanations for a sex ratio bias, stemming from male-male contests for matings, are discussed. We conclude that the effect of environmental conditions on adult sex ratio varies dramatically even in closely related and ecologically similar sympatric species. Therefore, difficulties in explaining the ecological predictors of sex ratio biases are likely due to high stochasticity rather than limited sample size.