Background Anaemia among small children in tropical Africa is common and often caused by infection with Plasmodium falciparum. The diagnosis of anaemia is difficult without a laboratory estimation of haemoglobin. The aim of this study was to examine if clinical findings related to malaria and anaemia would help to detect moderate and/or severe anaemia in children in rural Tanzania. Methods Children between 6 and 36 months were examined by health workers in an Out Patient Department (OPD) to detect severe anaemia (packed cell volume, PCV ≤ 20%) and in a cross sectional survey at village level to identify moderate anaemia (PCV 21–25%). History of recent fever and treatments was recorded and a clinical examination was performed. Findings In the survey, comparison of 65 moderately anaemic children with 373 mild/non anaemic children revealed no differences in history of fever or in the clinical examination. In the OPD comparison of 100 severely anaemic children with 116 non-severely anaemic control children revealed that pallor, respiratory rate, number of fever days last week, deteriorated general condition, heart rate, age, splenomegaly, low body weight and elevated body temperature were all indicators of severe ‘anaemia, only pallor, respiratory rate, fever days and palpable spleen however, remained associated with severe anaemia in multiple regression analysis. The combination of any pallor and either respiratory rate >55/min or fever >3 days, could predict severe anaemia with a sensitivity of 96% and a specificity of 71%. This was better than the currently recommended signs of severe pallor or an approximation of the Integrated Management of Childhood Illness (IMCI) criteria's for referral of children. Interpretation At primary health care level detection of severe anaemia can be improved by information about fever duration and determination of respiratory rate in children with pallor.