An approach allowing the on-target computation of efficiency and yield of large (hundreds of kW), grid-connected photovoltaic (PV) parks is presented. An efficiency metric is used to trigger alarms and to plan maintenance interventions. An accurate simulation model of the system, that is capable of taking into account all relevant environmental variables, has been used to precisely determine the expected baseline performances. Measured and simulated power together with string current data are constantly updated and monitored and then processed by using a novel figure of merit (FOM), to define aggregate efficiency indices both at the overall park level and at single string level. Aggregate indices on strings narrow down the search for sub-performing elements and allow the determination, possibly in one day, of whether the entire PV system or one of its components is not performing as expected. A workflow based on this approach is introduced along with some case studies attesting its effectiveness.