The problem of the behaviour of lumber at the low end of the strength distribution is slowly being unravelled. It appears that this lumber behaves more like a brittle material than had been previously assumed. This paper describes a test program in which matched samples (N = 88) of lumber were tested using different loading systems. It was found that the systems in which energy was stored in the testing machine gave apparent strengths that were lower than those obtained in a deformation-controlled setup. Since failure is caused by crack growth, it is postulated that the available stored energy is the cause of the observed strength differences. The difference between the two systems was found to be strength dependent (larger for the weaker pieces), and at the 5th percentile amounted to about 30%. The implications for the development of characteristic strength and design data are outlined.