Template relation construction as one phase in information extraction system aims to recognize a relation between two named entities. It recognizes relation between organization and position (organization-position), position and person (position person) and relation between person and organization (person-organization) that exist in a document. Experiments use 120 documents which were collected from the Indonesian online newspaper (www.kompas.com and www.republika.co.id). We use 60 documents as training documents and the rest as testing documents. Experiments show that threshold has no effect on results and adding more training documents give better accuracy. In general, experiments result show that recognition on organization-position relation gives the highest F-Measure. Keywords: Template Relation Construction, Information Extraction, Vector Space Model