Monitoring of gaseous releases from nuclear installations in the environment and air cleaning efficiency measurement are based on regular measurements of concentrations of contaminants in outlet chimneys and ventilation systems. The concentration distribution may be heterogeneous at the measuring point if the distance setting of the mixing is not sufficient. The question is about the set up of the measuring point in duct and the error compared to the homogeneous concentration in case of non-compliance with this distance. This study defines the so-called "well mixing length" from laboratory experiments. The bench designed for these tests allowed to reproduce flows in long circular and rectangular ducts, each including a bend. An optical measurement technique has been developed, calibrated and used to measure the concentration distribution of a tracer injected in the flow. The experimental results in cylindrical duct have validated an analytical model based on the convection-diffusion equation of a tracer, and allowed to propose models of good mixing length and representativeness of sampling points. In rectangular duct, the acquired measures constitute a first database on the evolution of the homogenization of a tracer, in the perspective of numerical simulations exploring more realistic conditions for measurements in situ.