It has been argued by Newman and Stein that, due to the chaotic size dependence, it may not be possible to take the infinite volume limit of a spin glass sample, and that one should introduce the metastate, a probability measure on Gibbs states.

I present a numerical construction of the metastate for the 3d Edwards-Anderson spin glass model, and discuss the results in the light of the so called non standard RSB picture of finite dimensional spin glasses.