Prediction of tensile properties by pulsed NMR measurements is improved by swelling the sample with a deuterated solvent. However, in recent years, the use of organic solvents has been avoided from the viewpoint of safety and health risks and the environmental impact of volatile organic compounds. In this study, we investigated a method for predicting the tensile properties of rubber by pulsed NMR measurements without using a swelling solvent. Spin-spin relaxation measurements in dry state and tensile tests were performed on sliced sheets sampled from the surface and interior of thermally oxidized thick-walled rubber. As a result of multivariate analysis using NMR parameters obtained from spin-spin relaxation, it was found that the spin-spin relaxation time of network component (T_2HS) and its Weibull coefficient (W_HS) are effective as explanatory variables for tensile properties. Regression equations using T_2HS and W_HS could predict the extension ratio at break (λ_B) and the stress at 100% strain (S₁₀₀) of surface and inner layers of thermally oxidized products with high accuracy.