This work describes human-like learning in Dynemotion agents. These agents perceive their world through the combined lenses of multi-layered emotion-based concepts (memories), personality traits, and current needs. They learn in a psychologically plausible and computationally tractable fashion that focuses on perceptions that have emotional significance. This method enables concept formation, aggregation, and generalization, and takes into account a working store and long-term memory fading with disuse. Agents may learn from observation, from their own actions, and from what other agents tell them. This leads to a form of reputation and opinion in learning as well as first-hand learning. Another emergent effect of this learning algorithm is that agents learn in ways that make them subject to human-like correlation/causation errors. Dynemotion agents are designed to become socially plausible in personality, emotion, behavior, learning, and relationships with each other and humans, as supported by their emotional and learning systems.