Introduction: Empathy is considered to be an important therapist offered condition. Historically the exploration of empathy has employed quantitative methodologies. It is argued that these methodologies cannot capture the socially constructed nature of psychological concepts and instead regard empathy as problematic due to its inconsistently applied definition. Objectives: This study aimed to explore therapists' discourse around empathy by employing a qualitative methodology and acknowledging the importance of context. A further objective was to encourage a theoretical and methodological shift in the way that psychological concepts are conceptualised and investigated. Design: A discursive psychological approach was taken in the analysis of data from discussion groups. Method: Discourse was collected from two discussion groups conducted at an NHS Primary Care Trust: the first with a group of clinical psychologists and the second with a group of cognitive behaviour therapists. In addition some documentary information was collected from the research site in order to contextualise the service. Results: In both discussion groups, empathy was considered fundamental to the therapeutic relationship between the client and therapist. Therapists constructed empathy in two ways: as a limited therapist experience and as a quality that might develop over time. Further patterns emerged in the data; the clinical psychologists made frequent use of case studies whereas the cognitive behavioural therapists cited research evidence and made use of theoretical models. Discussion: The results are discussed with reference to a particular model of discursive psychology where the activity done through discourse is emphasised. It is argued that through particular constructions of empathy, therapists were working up their professional accountability. It is suggested that therapists work up their constructions of empathy as factual and therefore indisputable through discursive devices, identified as the use of case studies and research evidence.