According to PricewaterhouseCooper's "Global Economic Crime Survey 2014", 37% of the 5,128 organizations surveyed had been victims of economic crime. Given this rise in economic crimes, investigators are working on novel forensic methods to detect financial fraud. Traditional data driven fraud detection methods are only capable of detecting transactions that correspond to already known fraud schemes. An open question is how to detect multi-dimensional patterns indicating 'anomalous' transactions in very large volumes of accounting data. This talk will present PricewaterhouseCooper's and the DFKI's recently developed approach to detect anomalous journal entries / transactions in large scale financial and accounting data by the use of deep learning and stacked autoencoders.