The approximations implicit in the use of the Gaussian network model for soft rubber are discussed. It is shown that the form of the stress–strain curve can be derived for this model simply, and without special assumptions about the form or behavior of the network. The common assumption that the network junctions are fixed, or can be treated as fixed, is discussed. It is shown that this picture of the situation is unrealistic: the junctions have a Brownian motion comparable to that of any portion of the intervening molecular segments. The introduction of this assumption is not generally admissible, but it will not affect the outcome of certain types of calculation; in particular, one can foresee that it need not affect the calculated form of the stress-strain curve. A particularly simple and straightforward calculation of the network entropy on this basis is given. Wall's theory of rubber is analysed. It is shown that Wall's postulates are not consistent with the network structure of rubber, and in general lead to different results.