The rest-frame intrinsic UV luminosity is often used as an indicator of the instantaneous star formation rate (SFR) in a galaxy. While it is in general a robust indicator of the ongoing star formation activity, the precise value of the calibration relating the UV luminosity to the SFR (Bν) is sensitive to various physical properties, such as the recent star formation and metal enrichment histories, along with the choice of stellar initial mass function (IMF). The distribution of these properties for the star-forming galaxy population then suggests that the adoption of a single calibration is not appropriate unless properly qualified with the uncertainties on the calibration. We investigate, with the aid of the galform semi-analytic model of galaxy formation, the distribution of UV-SFR calibrations obtained using realistic star formation and metal enrichment histories. At z = 0, we find that when the IMF is fixed (to the Kennicutt IMF), the median calibration is Bfuv = 0.9 where SFR/[M⊙ yr−1] = Bν × 10−28 × Lν/[erg s−1 Hz−1]. However, the width of the distribution Bfuv suggests that for a single object there is around a 20 per cent intrinsic uncertainty (at z = 0, rising to ≃30 per cent at z = 6) on the SFR inferred from the FUV luminosity without additional constraints on the star formation history or metallicity. We also find that the median value of the calibration Bfuv is correlated with the SFR and redshift (at z > 3) raising implications for the correct determination of the SFR from the UV.