This study characterizes the spatial dynamics of Plum Leaf Scald on three commercial orchards located in Sao Paulo State, Brazil. All trees from each orchard were evaluated during four consecutive years (one orchard), and for three consecutive years (two orchards). The incidence and characteristics of spatial patterns were assessed by the binomial dispersion index, Ripley's K-function and average minimum distance. Autologistic models were fitted for different neighboring structures with model selection guided by the Akaike Information Criteria. The autologistic model relates the probability of an infection with the presence of the disease on neighboring trees within and between the rows and also on the diagonals, on the same and/or previous time of assessment. Aggregated patterns were detected by different methods, except for the evaluations with incidence below 10%. The selected autologistic model consistently showed an increased risk of disease for trees with infected neighboring in the same row on the same time of assessment, whereas effects between rows were only detected for the higher incidences. Overall, the Plum Leaf Scald epidemic shows random patterns at the beginning and aggregation on later stages.