In this paper, we propose a new method of estimating *T*_{1} maps *retroactively* that only requires the acquisition or availability of four conventional MRI sequences. For these sequences, we employ a novel normalization method using cerebellar gray matter as a reference tissue, which allows diffuse differences in cerebral normal-appearing white matter (NAWM) to be detected. We use a regression model, fit separately to each tissue class, that relates the normalized intensities of each sequence to the acquired *qT*_{1} map value at each voxel using smooth functions. We test our model on a set of 63 subjects, including primary progressive (PPMS), relapsing-remitting (RRMS) and secondary progressive multiple sclerosis (SPMS) patients and healthy controls, and generate statistical *qT*_{1} maps using cross-validation. We find the estimation error of these maps to be similar to the measurement error of the acquired *qT*_{1} maps, and we find the prediction error of the statistical and acquired *qT*_{1} maps to be similar. Visually, the statistical *qT*_{1} maps are similar to but less noisy than the acquired *qT*_{1} maps. Nonparametric tests of group differences in NAWM relative to healthy controls show similar results whether acquired or statistical *qT*_{1} maps are used, but the statistical *qT*_{1} maps have more power to detect group differences than the acquired maps.