Rose Coloured Starlings ( Sturnus roseus) flew repeatedly for several hours in a wind tunnel while undergoing spontaneous variation in body mass. The treatments were as follows: flying unrestrained (U), with a control harness of 1.2% of their body mass (C), or with a harness of 7.4% of their body mass, which was either applied immediately before the flight (L S) or at least 9 days in advance (L L). Energy expenditure during flight (e f in W) was measured with the Doubly Labelled Water method. Flight costs in L S and L L were not significantly different and therefore were pooled (L). The harness itself did not affect e f, i.e. U and C flights were not different. e f was allometrically related with body mass m (in g). The slopes were not significantly different between the treatments, but e f was increased by 5.4% in L compared to C flights (log 10(e f) = 0.050 + 0.47 × log 10( m) for C, and log 10(e f) = 0.073 + 0.47 × log 10( m) for L). The difference in e f between C, L S and L L was best explained by taking the transported mass m transp (in g) instead of m into account (log 10(e f) = −0.08 + 0.54 × log 10( m transp)). Flight costs increased to a lesser extent than expected from interspecific allometric comparison or aerodynamic theory, regardless of whether the increase in mass occurred naturally or artificially. We did not observe an effect of treatment on breast muscle size and wingbeat frequency. We propose that the relatively low costs at a high mass are rather a consequence of immediate adjustments in physiology and/or flight behaviour than of long-term adaptations.