Using a method recently proposed by Sinai and Yakhot, we obtain the equation for the probability density function (pdf) of the temperature increments in fully developed turbulence. The closure problem is reduced to the determination of two conditional expectations and can be carried out mainly by requiring isotropy and a weak correlation of the physical variables. As a consequence of our analysis, we show that the pdf is nearly Gaussian about zero and then has a stretched exponential behavior for large fluctuations.