Abstract. The exact expression for the probability distribution function (pdjO, P(Au~), of a velocity difference AuT, over a distance r, in incompressible fluid turbulence, obtained from the Navier-Stokes equations, is used as a basis for deriving approximate profiles for P (Au~). These approximate forms are deduced from an approximate factorisation of the underlying fimctional probability distribution of the flow field, in which the individual factors capture different physical effects. P(AuT) is represented as the integral, with respect to the spatially averaged dissipation rate e~, of the product of the conditional pdf of Aur given e~, and the pdf of e~. The approximation yields the latter as a log-Poisson pdf, while the conditional pdf is found to be a Gaussian for a transverse increment, and the product of a Gaussian and a cubic polynomial for a longitudinal increment. This approximation is equivalent to the refined similarity hypothesis coupled with the log-Poisson distribution, and it possesses the characteristic features of P(Au~), including symmetric profiles for transverse increments, asymmetric profiles for longitudinal increments, and the development of pronounced non-Gaussian features at small separations. The associated scaling exponents for longitudinal and transverse structure functions are shown to be identical, in this approximation, and to assume the log-Poiss0 n form.