The prognostic value of serum ferritin (SF) as a predictor of bone marrow iron stores (BMS) was evaluated in an attempt to use SF and other readily available clinical and laboratory parameters in the more accurate prediction of BMS without bone marrow aspiration. The results of routine haematological studies of 1000 patients referred for a bone marrow examination as part of their diagnostic evaluation at Tygerberg Hospital were statistically analysed. Significant associations were found. The CHAID programme of automatic interaction detection as developed by Kass was used to determine the optimum classification. A statistical model in the form of a dendrogram has been computed for this hospital population and this laboratory for future predictions of BMS using only the values for SF and mean corpuscular volume. The predictions were statistically very highly significant, especially in patients with iron deficiencies.