Objective. To identify, in conservatively treated, very early arthritis patients, predictors of >1 erosion(s) at 2 years, and to construct a prediction model. Methods. Community-based adults (n ؍ 310) who had never taken disease-modifying antirheumatic drugs (DMARDs) or steroids with swelling of >2 joints persisting for >4 weeks and lasting <6 months were recruited. Erosion status was assessed at 0, 6, 12, and 24 months; evaluations were comprised of clinical criteria (Disease Activity Score, Health Assessment Questionnaire), C-reactive protein level, erythrocyte sedimentation rate, autoantibodies, bone and cartilage markers, hand densitometry, and HLA class II shared epitopes. Patients meeting American College of Rheumatology rheumatoid arthritis (RA) criteria or with undifferentiated arthritis (UA) were followed and treated conservatively: one-third of RA patients and three-fourths of UA patients received no DMARDs during 2 years; a biologic agent was given to 1.8% of the patients during the first year. The main judgment criterion was >1 erosion(s) at 2 years. Results. At 2 years, 219 patients were assessed; 31.3% with RA and 10.6% with UA had >1 erosion(s). Logistic regression analysis at that time showed erosion(s) strongly associated with serum IgA rheumatoid factor (IgA-RF) and pyridinoline levels for the 190 patients with no baseline erosions, with the corresponding receiver operating characteristic curve having an area under the curve of 0.77 (95% confidence interval 0.64 -0.86). A prediction model was constructed with IgA-RF thresholds of 5 and 25 IU/ml and a pyridinoline threshold of 10 nM/liter; odds ratios ranged from 1 for IgA-RF <5 IU/ml and pyridinoline <10 nM/liter to 50.75 for the association of IgA-RF >5 IU/ml and pyridinoline >10 nM/liter. Conclusion. This model, using serum IgA-RF and pyridinoline concentrations, was able to predict >1 erosion(s) at 2 years in very early arthritis patients.