Uncertainties related to the material properties of a composite material can be determined from the micro-, meso-or macro-scales. These three starting points for a stochastic modelling of the material properties are investigated. The uncertainties are divided into physical, model, statistical and measurement uncertainties which are introduced on the different scales. Typically, these uncertainties are taken into account in the design process using characteristic values and partial safety factors specified in a design standard. The value of the partial safety factors should reflect a reasonable balance between risk of failure and cost of the structure. Consideration related to calibration of partial safety factors for composite material is described, including the probability of failure, format for the partial safety factor method and weight factors for different load cases. In a numerical example, it is demonstrated how probabilistic models for the material properties formulated on micro-scale can be calibrated using tests on the meso-and macro-scales. The results are compared to probabilistic models estimated directly from tests on the macro-scale. In another example, partial safety factors for application to wind turbine blades are calibrated for two typical lay-ups using a large number of load cases and ratios between the aerodynamic forces and the inertia forces.