We are interested in the manipulation skills required by future service robots performing everyday tasks such as preparing food and cleaning in a typical home environment. Such robots must have a robust set of skills that can he applied in the unpredictable and varying circumstances that arise in everyday life. To succeed in such a setting, a service robot must have a strong ability to transfer old skills to new varied settings. We are inspired by the strong transfer ability demonstrated by infants and toddlers on simple manipulation activities, and we are motivated to try and replicate these abilities in an artificial system. We treat this as a problem of making analogies, and describe a theoretical framework which could account for it. We sketch the ideas of a computational model for implementing the required analogical reasoning.