Several quantitative MRI metrics have been proposed to quantify myelin in the central nervous system but each of them includes confounding factors that impair their sensitivity and specificity. Because these factors are different across metrics, data driven approaches developed for blind source separation problems to extract the common component between recordings of the same sources seem appropriate. This study compares linear and nonlinear methods to combine myelin-sensitive metrics: T1, MTR, MTsat, MTV (1 – PD). The repeatability of the resulting combined metrics as well as their sensitivity to different microstructural features are tested. A higher sensitivity is achieved with linear combinations.