The neuroscience of empathy has enormously expanded in the past two decades, thereby making instrumental progress for the understanding of neural substrates involved in affective and cognitive aspects of empathy. Yet, these conclusions have relied on ultrasimplified tasks resulting in the affective/cognitive dichotomy that was often modeled and overemphasized in pathological, developmental, and genetic studies of empathy. As such, the affective/cognitive model of empathy could not straightforwardly accommodate and explain the recent surge of neuroscientific data obtained from studies employing naturalistic approaches and intergroup conditions. Inspired by phenomenological philosophy, this article paves the way for a new scientific perspective on empathy that breaks thorough the affective/cognitive dichotomy. This neuro-phenomenological account leans on phenomenological analyses and can straightforwardly explain recent neuroscience data. It emphasizes the dynamic, subjective, and piecemeal features of empathic experiences and unpicks the graded nature of empathy. The graded empathy hypothesis postulates that attending to others' expressions always facilitates empathy, but the parametric modulation in the levels of the empathic experience varies as a function of one's social interest (e.g., via intergroup or inter-personal cues) in the observed other. Drawing on multiple resources that integrate neuroscience with phenomenology, we describe the potential of this graded framework in an era of real-life experimentation. By wearing lenses of neuro-phenomenology, this original perspective can change the way empathy is considered.