The feasibility to implement fast-chemistry reactions in a three-dimensional large eddy simulation (LES) of a turbulent reacting flow using a filtered density function (PDF) technique is studied. Low-density polyethylene (LDPE) is used as an representative reaction due to the stiff nature of the ordinary differential equation (ODE's) describing the kinetics. In FDF/LES, the chemistry needs to be evaluated many times for a large number of fictitious particles that are tracked in the flow, and therefore a constraint is put to the CPU time needed to solve the kinetics. Pope (1997) developed an in situ adaptive tabulation (ISAT) to treat complex chemistry computationally very efficiently when many evaluations of the chemistry are needed. Kolhapure and Fox (1999) successfully applied IS AT to LDPE using a quasi steady state assumption (QSSA). In the present paper, the aim is to optimize the latest version (ISAT Version 4.0, Pope, 2003) for the full LDPE reaction (i.e. without QSSA), in terms of accuracy and speed up by varying the error tolerance and the number of trees used by ISAT. For this purpose, a pairwise mixing stirred reactor (PMSR) is employed, since it forms a stringent test for the chemistry solver due to the large accessed region of composition space that can be established. For a number of trees of Ntree = 8 and an error tolerance of εtol = 10-5 the best overall performance of ISAT was obtained: compared with direct integration, a speed up factor of more than ten combined with an relative error in temperature of about 1% was found.