The search for applicable and valid research evidence-based practice articles is not supported well in common EBP resources, as some crucial study data, such as patient details, study design and results, are not available or presented explicitly. We propose to extract these data from research articles using a two-step supervised soft classification method. Compared to manual annotation, our approach is less labor-intensive and more flexible, hence opening up the possibility of utilizing these data to facilitate the evidence selection process in information seeking support systems.