OBJECTIVE To measure the reliability of the Mini Nutritional Assessment (MNA) in institutionalized elderly people. METHODS 12 day interobserver reliability study. METHODS All subjects admitted to two long term geriatric units in Mataró (Barcelona, Spain) over 4 months during 1996 (n=67). METHODS in each center, different trained nurses independently administered the MNA on two separate occasions. RESULTS Mean (standard deviation) scores for the two assessments of the MNA were 20.8 (5.4) and 21.3 (4.6) respectively. Internal consistency, estimated by the Cronbach's Alpha, were 0.83 and 0.74 for the first and second assessment respectively. Test-retest reliability, according to the intraclass correlation coefficient (ICC), was 0.89 for the total MNA score and higher than 0.89 for its continuous items. According to the Kappa index, test-retest reliability for the stratified total MNA was substantial (0.78); for the 18 ordinal or nominal items of the MNA it was 'almost perfect' or 'substantial' in 12 items, 5 were 'moderate' to 'fair' and in I item it was 'slight'. Subjective health evaluation, the number of glasses of liquids per day, and brachial circumference (this former with an ICC=0.91) were the items with the lowest Kappa indices. CONCLUSIONS The MNA test has good levels of reliability, according to its internal consistency and its test-retest reproducibility. Some improvements can still be introduced by refining the categorization and content of some items with low reliability.