Agricultural research conducted on-farm poses significant challenges. There is the inherent complexity of a farm business, the variability of the environment, and often a mismatch in scale between experimentation and farm practice. These issues are also motivating factors in the recent resurgence of interest in moving research on to the farm. Research carried out on-farm and in close collaboration with farmers can help ensure that knowledge generated is relevant to today’s multifaceted farm operation. To assist in coping with complexity, a number of on-farm trial designs will be reviewed. Which trial design is the right one will depend on research objectives. Basic research evaluating biological and chemical processes is different than adaptive research and developing recommendations. Different experimental designs are required, although I contend that all benefit from being conducted in a ‘real world’ farm environment. Basic research trials will generally be carried out over the long-term, with researcher designed treatments, intensive instrumentation and multiple replications at each farm site. The number of on-farm sites will often be constrained. Adaptive research will generally be dynamic, with farmer input into treatments, and an iterative process of design and redesign carried out over time. Replications often occur through including a large number of on-farm sites, with few or no replications within each farm site. Both types of trials can be linked together in a systematic manner. This is called the ‘mother daughter’ trial design, which involves a central mother trial at  a large farm or research station farm, with all treatments represented and within-site replication, linked to numerous daughter trials that test a subset of technologies, often with one-replicate per farm site. A reference plot is often a feature incorporated to evaluate farm to farm variation, along with site-specific, farmer-designed treatments as another type of control. In addition, a standardized control is included across sites. Geo-positioning technologies can be used to help understand and document complexity. Socio-economic instruments and multidisciplinary collaboration is also important, to document farmer preferences and evaluation of technologies.