In excess of 2,100 invasive breast cancers, which had been detected through screening mammography, were investigated to determine factors that were predictive of lymph node status.  Multivariable models indicated that lymph nodes were more likely to show cancer if the primary lesions were larger, the histological type was either infiltrating ductal or lobular, and the cancer was multifocal and palpable. Younger women were more likely to show nodal involvement. Also nodal involvement was more common when an extensive in situ component (EIC) was reported. Tumour grade was predictive of nodal spread as a univariate predictor. It would be more accurate if screening assessment clinics were to use models to predict nodal status, which were developed from their own clinical experience, rather than to use generic models developed in other settings that mostly related to symptomatic cancer. These models could assist clinical decision-making on axillary node dissection and also provide guidance to pathologists on numbers of tissue sections to examine. Models are reported for predicting nodal involvement in women whose cancers were detected in an Australian screening service. A practical scenario is described where such models could be used to triage cancers according to predicted probability of nodal involvement, such that patients could receive either a non-invasive axillary investigation, a sentinel node biopsy, or an axillary clearance.