Existing anomaly detection methods are sensitive to units and scales of measurement. Their performances vary significantly if feature values are measured in different units or scales. In many data mining applications, units and scales of feature values may not be known. This paper introduces a new anomaly detection technique using unsupervised stochastic forest, called ‘usfAD’, which is robust to units and scales of measurement. Empirical results show that it produces more consistent results than five state-of-the-art anomaly detection techniques across a wide range of synthetic and benchmark datasets.