Predictable behaviour (or “behavioural stability”) might be favoured in certain ecological contexts, e.g. when representing a quality signal. Costs associated with producing stable phenotypes imply selection should favour plasticity in stability when beneficial. Repeatable among-individual differences in degree of stability are simultaneously expected if individuals differ in ability to pay these costs, or in how they resolve cost-benefit trade-offs. Bird song represents a prime example, where stability may be costly yet beneficial when stable singing is a quality signal favoured by sexual selection. Assuming energetic costs, ecological variation (e.g. in food availability) should result in both within- and among-individual variation in stability. If song stability represents a quality signal, we expect directional selection favouring stable singers. For a three-year period, we monitored 12 nest box plots of great tits Parus major during breeding. We recorded male songs during simulated territory intrusions, twice during their mate’s laying stage, and twice during incubation. Each preceding winter, we manipulated food availability. Assuming that stability is costly, we expected food-supplemented males to sing more stable songs. We also expected males to sing more stable songs early in the breeding season (when paternity is not decided), and stable singers to have increased reproductive success. We found strong support for plasticity in stability for two key song characteristics: minimum frequency and phrase length. Males were plastic because they became more stable over the season, contrary to expectations. Food-supplementation did not affect body condition but increased stability in minimum frequency. This treatment effect occurred only in one year, implying that food supplementation affected stability only in interaction with (unknown) year-specific ecological factors. We found no support for directional, correlational, or fluctuating selection on the stability in minimum frequency (i.e., the song trait whose stability exhibited cross-year repeatability): stable singers did not have higher reproductive success. Our findings imply that stability in minimum frequency is not a fitness quality indicator unless males enjoy fitness benefits via pathways not studied here. Future studies should thus address the mechanisms shaping and maintaining individual repeatability of song stability in the wild.