Aim: To test long standing theory on the role of environmental conditions (both mean and predictability) in shaping global patterns in the egg sizes of marine fishes. Location: Global (50˚ S to 50˚ N). Time period: 1880 to 2015. Major taxa studied: Marine fish. Methods: We compiled the largest geo-located dataset of marine fish egg size (diameter) to date (n = 1,078 observations; 192 studies; 288 species; 242 localities). We decomposed sea surface temperature (SST) and chlorophyll-a time-series into mean and predictability (seasonality and colour of environmental noise – i.e. how predictable the environment is between consecutive time steps), and used these as predictors of egg size in a Bayesian phylogenetic hierarchical model. We test four specific hypotheses based on the classic discussion by Rass (1941), as well as contemporary life-history theory, and the conceptual model of Winemiller and Rose (1992). Results: Both environmental mean and predictability correlated with egg size. Our parsimonious model indicated that egg size decreases by c. 2.0-fold moving from 1˚C to 30˚C. Environments that were more seasonal with respect to temperature were associated with larger eggs. Increasing mean chlorophyll-a, from 0.1 to 1 mg m^-3^, was associated with a c. 1.3-fold decrease in egg size. Lower chlorophyll-a seasonality and reddened noise were also associated with larger egg sizes – aseasonal but more temporally autocorrelated resource regimes favoured larger eggs. Main conclusions: Our findings support results from Rass (1941) and some predictions from Winemiller and Rose (1992). The effects of environmental means and predictability on marine fish egg size are largely consistent with those observed in marine invertebrates with feeding larvae, suggesting important commonalities in how ectotherm egg size responds to environmental change. Our results further suggest that anthropogenically-mediated changes in the environment will have profound effects on the distribution of marine life histories.