In this paper, we solve approximately a physical facility layout by using two genetic algorithm approaches. These approaches build either feasible solutions or infeasible solutions using penalization in the objective function for the infeasible ones. A particular type of encoding is used for each approach. Two types of encoding are formulated : a direct encoding for wich specific mutation and crossover operator must be designed as well as a mixed encoding where a function (placement algorithm in our case) builds efficient solutions. We consider a single material handling system moving product parts between cells in an Euclidean distance. The workshop contain an input and an output. It space is encumbered with obstacles such as pillars and walls. Cells are linked by distance constraints : a minimal and a maximal distance must be respected. These cells have different shapes. Our approaches are compared to a simulated annealing approach. Some numerical examples issued from large size problems are efficiently solved.