Current settings for concentrate allocation and milking frequency per cow are based on knowledge about the overall population mean. Within herds there is considerable variation in feed efficiency and milking characteristics, both between cows and within cows over time (Duinkerken et al. 2003; 2006). So far this variation is ignored and so current settings are suboptimal. Allowing for individual and temporal variation in feed efficiency and milking characteristics can increase economic results of dairy production. To realise this increase in economic results, knowledge about the actual individual response in milk production to changes in concentrate allocation and milking frequency is needed (Wathes et al., 2005). Actual responses can be estimated from recently acquired individual process data by use of an adaptive model. Currently, adaptive models have some disadvantages when applied to biological processes in a complex dynamic environment. Probably, these problems can be overcome by use of dynamic linear models (DLM), developed by West and Harrison (1997). The objectives of this research are: 1. To quantify individual and temporal variation in dairy cow feed and milking efficiency and to identify adjustable inputs in order to assess economic prospects. 2. To compare different adaptive models for on-line parameter estimation, with particular attention to dynamic linear models (DLM). 3. To develop a DLM for on-line parameter estimation, involving individual feed efficiency and milking performance. 4. To develop a decision support system (DSS) for computer control of milking frequency and concentrate allocation. 5. To test, improve and validate the DSS on a research farm. 6. To discuss benefits and consequences of the DSS for interrelated farm processes.