The current study builds upon earlier work by Baheti and Scott (1980), Khalsa et al (2011) and Pye and Zhang (2012), to analyze the process of automated misalignment correction with the introduction of an uncertainty analysis applied to an experimental training data set. The accuracy of correction process from this experimental data is quantified, allowing a criterion to be applied to determine whether or not sufficient training has been completed for each heliostat to mean overall field accuracy requirements. To investigate the potential improvements from extended training, a synthetic data set is generated, and used to investigate preferred times of year and times of data for training specific heliostats in the field. Summer data is shown to be best, but the additional of some winter data is helpful. Time-of-day is also important, especially for the sides of the heliostat field; middle-of- the-day training and spring or autumn training are seen to be less effective. A training programme for the entire heliostat field is presented and discussed: each heliostat is trained daily in summer for two minutes, and daily in winter for one minute in the morning and evening, resulting in 95% certainty that all heliostats will have their focal spot within 1.5 m of the target for the entire year, by an entirely automated process.