Aim. The present study evaluated the reliability of The Cohen-Mansfield Agitation Inventory Polish version among demented and non-demented nursing home residents. Method. 35 nursing home residents fulfilling the ICD-10 criteria for dementia and 32 non-demented residents were assessed with The Cohen-Mansfield Agitation Inventory Polish version. For analysis of data we used Cronbach's alpha and Spearman's rank correlation coefficients. Results. The internal consistency of the CMAI and interrater reliabilities were satisfactory both for dementia and the control groups. Although certain differences between the demented and non-demented residents were reported. Conclusion. The preliminary result from the evaluation of the reliability of The Cohen-Mansfield Agitation Inventory Polish version were all satisfactory, both for demented and non-demented residents.