Modelled transport equations for the conditional mean and r.m.s. temperature increments theta before autoignition in a turbulent non-premixed flow have been developed. The model is based on the recent finding that a well-defined mixture fraction, f(MR), first ignites and the concept that the conditional scalar dissipation rate, chi\f(MR), controls the hear losses and hence the ignition time. By introducing the fluctuations of chi\f(MR) and by modelling the conditional correlation coefficients between chi and theta from results from Direct Numerical Simulations, the model is closed and calculates autoignition times in good agreement with DNS results. Extensions to complex chemistry and the relationship with the more rigorous Conditional Moment Closure are discussed. Finally, the model is incorporated in a Favre-averaged k - epsilon code (KIVA-II) to predict autoignition of a transient fuel jet in hot air with promising results.