Open access. Mechanisms by which liaison mental health services (LMHS) may bring about improved patient and organisational outcomes are poorly understood. A small number of logic models have been developed, but they fail to capture the complexity of clinical practice. We synthesised data from a variety of sources including a large national survey, 73 in-depth interviews with acute and liaison staff working in hospitals with different types of liaison mental health services, and relevant local, national and international literature. We generated logic models for two common performance indicators used to assess organisational outcomes for LMHS: response times in the emergency department and hospital length of stay for people with mental health problems. We identified 8 areas of complexity that influence performance, and 6 trade-offs which drove the models in different directions depending upon the balance of the trade-off. The logic models we developed could only be captured by consideration of more than one pass through the system, the complexity in which they operated, and the trade-offs that occurred. Our findings are important for commissioners of liaison services. Reliance on simple target setting may result in services that are unbalanced and not patient-centred. Targets need to be reviewed on a regular basis, together with other data that reflect the wider impact of the service, and any external changes in the system that affect the performance of LMHS, which are beyond their control.