The relative impact of selection, chance and history will determine the predictability of evolution. There is a lack of empirical research on this subject, particularly in sexual organisms. Here we use experimental evolution to test the predictability of evolution. We analyse the real-time evolution of Drosophila subobscura populations derived from contrasting European latitudes placed in a novel laboratory environment. Each natural population was sampled twice within a three-year interval. We study evolutionary responses at both phenotypic (life-history, morphological and physiological traits) and karyotypic levels for around 30 generations of laboratory culture. Our results show (1) repeatable historical effects between years in the initial state, at both phenotypic and karyotypic levels; (2) predictable phenotypic evolution with general convergence except for body size; and (3) unpredictable karyotypic evolution. We conclude that the predictability of evolution is contingent on the trait and level of organization, highlighting the importance of studying multiple biological levels with respect to evolutionary patterns.