Ed Lorenz introduced idea the chaotic attractor. Whether the atmosphere has an attractor or not is probably not a useful question to answer. Forecast models, however, do have attractors, or attracting manifolds. In this paper we demonstrate using the Navy Operational Global Atmospheric Prediction System (NOGAPS) that the model attractor plays an important role in predictability. In particular, we will reveal using shadowing techniques how "initial condition" error and model error are related to the attractor and how these effect forecasts.