Compassion is potentially an effective emotion-regulation strategy to face the suffering of self and others. The aim of this paper is to provide an evolutionary understanding of compassion and compassion training (CT) by examining the psychological, neural and behavioral effects of loving-kindness meditation and compassion meditation. The author presents various definitions of compassion and examines the physiological and neural processes behind it. Compassion seems to have evolutionary roots but can be limited due to inherited blocks and fears. Compassion is however trainable and can potentially bypass certain evolutionary-based biases. CT results in various significant psychological effects, most notably positive affect, increased (self) compassion, and mindfulness. Evidence is however inconsistent, especially in relation to active controls. Neural effects are significant yet inconsistent across different experimental conditions. CT without a concurrent task activates (1) the right somatosensory cortices (2) the parieto-occipital sulcus, and (3) the right anterior insula. In relation to the socio-affective video task, CT activates medial orbitofrontal cortex, ventral striatum/nucleus accumbens, putamen, and anterior parts of anterior cingulate cortex; regions related to positive affect, motivational reward and affiliation. These findings converge with the reviewed psychological literature. CT also results in increased altruistic and compassionate behavior towards others, even when it’s costly to the self and under no-reciprocity conditions. Behavioral effects are mostly demonstrated in game-settings against active controls but also in one real-life situation. Together, the results suggest that CT is beneficial to individuals as well as inter-group relationships.