This is the public Matlab implementation for medical image soft-segmentation using the atlas-based expected label value (ELV) approach proposed by Aganj and Fischl (IEEE TMI 2021; IEEE ISBI 2019). This approach considers the probability of all possible atlas-to-image transformations and computes the ELV, without relying only on the transformation chosen as "optimal" by a registration method. This is done without deformable registration, thereby avoiding the associated computational costs. A short tutorial is included in EXAMPLE.m.