Probability density function (PDF) method is implemented in direct numerical simulation (DNS) to simulate turbulent reactive flows (DNS-PDF method). In the DNS-PDF method, a flow field and a non reactive scalar are predicted by the DNS, whereas reactive scalars are predicted by the Lagrangian PDF method, in which a transport equation of joint PDF of reactive scalars is solved by using a large number of notional particles. A mixing time scale for a mixing model used in the PDF method is directly estimated from the DNS result. In the present model for the mixing time scale, the effect of distance between notional particles is implicitly taken into account. The DNS-PDF method is applied to a planar jet with a second-order chemical reaction. The results show that the DNS-PDF method can accurately predict the rms value of mixture fraction fluctuation, and the present model for the mixing time scale is valid. It is also found that the DNS-PDF method can accurately predict mean concentrations of reactive species.