Traditionally, the accuracy of solar power forecasts has been measured in terms of classic metrics, such as root mean square error (RMSE) or mean absolute error (MAE), and it is widely accepted that the smaller the error, the greater the economic benefits. Nevertheless, this is not as straightforward as it may seem, because market conditions must be studied first. Relationships between magnitudes of deviations between forecast and actual production and market penalties that apply at each moment are crucial. In this study, we analyze various day-ahead production forecasts for a 1.86 MW photovoltaic plant considering different techniques and sets of inputs. A nRMSE of 22.54% was obtained for a Support Vector Regression model trained by numerical weather predictions (NWP). This model produced the most benefits. An annual forecasting value of 4788 with respect to a persistence model was obtained for trading in the Iberian (Spain and Portugal) day-ahead electricity market. Annual value added by the NWP service totaled 2801 and room for improvement regarding NWP variables rose to 3877. As a general trend, it was found that smaller errors (RMSE) generated higher incomes. For each 1 kW h improvement in RMSE, the annual value of forecasting increased 22.32. Nevertheless, some models that gave larger errors than others also brought greater benefits. Thus, market conditions must be considered to accurately evaluate model economic performance. © 2017 Elsevier Ltd