Some bacterial lineages seem to have undergone significant genome shrinkage over the last millions of years, a process called reductive evolution. For example, the genome of Buchnera aphidicola is only one-seventh the size of the genome of its close relative Escherichia coli (Moran and Mira, Genome Biol. 2001). This reductive evolution was initially thought to be a signature of intracellular lifestyle. Thus, explanatory mechanisms related to this lifestyle were proposed, like a smaller effective population size, a lower exposure to horizontal transfer, or the uselessness of some genes. However, reduced genomes were also found in some free-living cyanobacteria like Prochlorococcus marinus (Rocap et al., Nature 2003) or Pelagibacter ubique (Giovannoni et al., Science 2005). This questions the evolutionary mechanisms underlying reductive evolution: Are they shared by endosymbionts and cyanobacteria? Predicting the effect of these mechanisms would shed light on the minimal combination of factors required to observe reductive evolution. We performed in silico evolutionary experiments to test the effect of a smaller population size, a higher mutation rate, a less demanding environment, and a less varying environment. We used the individual-based model aevol (Knibbe et al., Mol. Biol. Evol. 2007), where virtual genomes undergo small-scale mutations, rearrangements, and reproduce differentially based on their performance at a curve-fitting task. Experiments began with ancestral populations that evolved with standard parameters during 150,000 generations. Then the value of the tested parameter was changed and evolution continued for 150,000 additional generations. These simulations showed that a higher mutation rate, as well as a less demanding environment, led to both fewer genes and shorter intergenic sequences. Environmental stabilization triggered reduction in non-coding sequences but not in gene number. Finally, with a smaller population size, some genes were lost but the genome expanded because intergenic sequences lengthened, even with an enforced deletional bias. To explain this surprising effect, one hypothesis is that the sudden reduction in population size leaves a less fit population. The selection, now directional, indirectly favors the most evolvable genomes, that is, genomes with longer non-coding sequences undergoing larger rearrangements. This study is thus a first step towards predicting the genome structure expected for a given evolutionary scenario. The next steps are to (i) further analyze the evolutionary trajectories in these experiments, (ii) test the effect of combined scenarii with multiple changed parameters, and (iii) find the best scenarii for intracellular and free-living species with reduced genomes.