Abstract In static and dynamic reservoir modeling, results have shown that the quality and accuracy of the models are dependent on the quality of rock and fluid property inputs e.g. facies, porosity, hydrocarbon saturation, permeability, net sand etc. used in creating the models. Electrofacies (facies defined using electric well logs) are a key property, since most other properties demonstrate some trend with facies. The typical workflow in modeling electrofacies uses well logs and some type of supervised or hierarchical (unsupervised) technique to cluster/ index the data into a number of groups (electrofacies). The input well logs however tend to ‘correlate’ or ‘anti-correlate’ (sometimes very strongly) and do not truly form a set of ‘independent’ variables that ‘sense’ different attributes of the facies. This paper presents a statistical technique for identifying and assigning electrofacies using a multivariate analysis technique (Principal Component Analysis (PCA)) to create ‘eigen-variables’ from well logs. These eigen-variables contain all the original information in the well-logs and, in contrast to the well logs, are unique and ‘linearly independent’; therefore resulting in log signatures that are much better used in predicting electrofacies. We have used this technique to model electrofacies in the Bonga Deepwater field, and the results presented in this paper show a very robust facies network (based on ‘Indexation Phase’ Self-Organizing Maps - SOM) with a higher predictive accuracy than achieved using the well logs directly.