Collaborative filtering algorithms need enough data to provide accurate and reliable predictions. Hence, two e-commerce sites holding insufficient data may want to provide predictions on their partitioned data with privacy. Different privacy-preserving collaborative filtering systems have been proposed for this purpose. Some attacks can be employed against such systems to derive confidential data. In this paper, attack scenarios are designed against horizontally and vertically partitioned data-based collaborative filtering with privacy schemes to show how much data can be derived. Also, how additional knowledge about the system helps data reconstruction is studied. Empirical outcomes on real data sets show that it is possible to derive high amount of private data in some cases. However, when there is no additional information and data is dense, data reconstruction success becomes very low.