Temporal environmental variation in nature includes a large component of random fluctuations, the magnitude and predictability of which are modified under current climate change. These fluctuations are expected to have large eco-evolutionary impacts on natural populations, mediated by plastic and evolutionary responses to changing environments. However, understanding and predicting these responses is still hampered by lack of strong experimental evidence. The aim of FluctEvol was to shed new light on population responses to stochastic environments and facilitating their prediction, by using a unique combination of approaches. First, we designed and analysed theoretical models of evolution and demography under a randomly changing optimum phenotype, producing new quantitative predictions. Second, we performed cross-species analyses of long-term datasets from natural populations, to estimate the prevalence and magnitude of fluctuating selection in the wild. And third, we used large-scale and automated experimental evolution in stochastic salinity with the micro-alga Dunaliella salina, an extremophile that thrives at high and variable salinities. We manipulated the predictability of fluctuations in salinity, and used high-throughput phenotyping to analyse population growth and extinction risk, as well as evolution of plasticity at multiple levels of the phenotype. Our model organism D. salina combined the benefits of experimental evolution in microbes (short generations, ample replication) with a priori knowledge of ecologically relevant adaptive traits, allowing for hypothesis-driven experiments. Our results have shown that natural selection varies substantially over time in the wild, and that temporal autocorrelation in the environment is likely to have crucial effects on extinction risk, evolution of plasticity, and the dynamics of population genetic change.