The enthalpy relaxation of a partially cured (70%) epoxy resin, derived from diglycidyl ether of bisphenol-A cured by methyl-tetrahydrophthalic anhydride with accelerator, has been investigated. The key parameters of the structural relaxation (the apparent activation energy Delta h*, the nonlinearity parameter x, and the nonexponentiality parameter beta) are compared with those of the fully cured epoxy resin. The aging rates, characterized by the dependences of the enthalpy loss and peak temperature on log(annealing time), are greater in the partially cured epoxy than they are in the fully cured resin at an equivalent aging temperature (T-a = T-g - 20 degrees C). There is a significant reduction in Delta h*, from 1100 kJ mol(-1) for the fully cured system to 615 kJ mol(-1), as the degree of cure is reduced. The parameter x determined by the peak-shift method appears essentially independent of the degree of cure (x = 0.41 +/- 0.03 for the partially cured resin compared with 0.42 +/- 0.03 obtained previously for the fully cured resin), and does not follow the usually observed correlation of increasing x as Delta h* decreases. This invariability of the parameter x seems to indicate that it is determined essentially by the local chemical structure of the backbone chain, and rather little by the supramolecular structure. On the other hand, the estimated nonexponentiality parameter beta lies between 0.3 and 0.456, which is significantly lower than in the fully cured epoxy (beta approximate to 0.5), indicative of a broadening of the distribution of relaxation times as the degree of cross-linking is reduced. Like the parameter x, this also does not follow the usual correlation with Delta h*. These results are discussed in the framework of strong and fragile behavior of glass-forming systems, but it is difficult to reconcile these results in any simple way with the concept of strength and fragility.