In this work, a method is presented for estimating the likely economic benefit from installing a steady state on-line optimizer on a plant. The objective of an online optimization scheme is to track the real process optimum as it changes with time. This must be achieved while allowing for disturbances to the process, ensuring process constraints are not violated. The benefit gained depends on the structure of the on-line optimizer. This structure includes the measurements used for the estimation of model parameters, parameters estimated in the model and the variables which are used as set points for passing the optimization results to the regulatory control level of the process. Using the estimate of the economic benefit, the "best" structure of the optimizer can be determined. The disturbances to the process have be described by both statistical and deterministic means. With a given disturbance description and set of structural decisions, an average economic return for the process with on-line optimization can be estimated. This average is found using a second order Taylor series expansion of the non-linear process model at a nominal operating condition. The average economic return of the process can be directly traded off against the cost of the necessary equipment for installing a particular on-line optimizer (i.e. instrument costs). Two sets of examples are presented. In the first set of examples all of the process model structures can be captured using the second order Taylor series expansion. These examples are used to demonstrate the different features of the analysis of an on-line optimization structure. The second set of examples demonstrates the analysis on a model of the Williams-Otto plant. This case study is used to test the procedure on a non-linear case study. The results generated are compared against Monte Carlo simulations of the non-linear process. Finally there is a discussion and summary of the conclusions from examples and suggestions for potential areas for further research.