It has been known for more than 50 years that the distribution of red blood cells (RBCs) in microvessel networks is highly heterogeneous. Yet, the phase separation effect (PS), i.e the non-proportional distribution of RBCs between the two daughter branches of a simple divergent bifurcation, is still poorly understood especially when RBC concentration (tube hematocrit) reaches high values (> 20%) and when vessel diameter is close to that of RBCs' (~8µm), as often encountered in microcirculation. The most commonly used PS law has been empirically derived in vivo by Pries et al. [1] but its validity has been recently questioned [2]. Therefore, experimental data obtained in controlled conditions are needed. We have designed squared micro-channel bifurcations, with side ranging from 5 to 20 microns, and adapted the injection geometry. Associated to a new method for the measurement of high tube hematocrit values, this allows for an in vitro systematic investigation of the parameters influencing the PS (tube hematocrit, confinement, bifurcation geometry). Our results are compared with Pries' PS law.