In this letter we suggest that a method recently proposed by Wayland et al. [1] for recognising determinism in a time series can also be used as a diagnostic to determine the minimum embedding dimension required for reconstructing phase space. Furthermore we suggest a simple extension to the Wayland et al. method which enables phase space to be reconstructed from input-output time series data. We compare the results of this extension to the results given by the extensions to the method of false nearest neighbours put forward by Rhodes and Morari [2] and the method of averaged false nearest neighbours by Cao et al. [3].