Uterine-contraction detection is a fundamental component of pregnancy monitoring. Electrohysterography (EHG) provides a non-invasive and accurate alternative to intrauterine pressure (IUP) measurements, and several techniques provide an estimated IUP (eIUP) based on the EHG alone. Commonly, EHG contraction detection is based on amplitude thresholding of the eIUP. We aim at improving the reliability of contraction detection, such that automatic contraction detection can be realized. An algorithm for template-matching of the eIUP signal is proposed. This method is based on Bayesian evidence using a Gaussian likelihood function to classify uterine activity. Gaussian templates are matched to the input signal, with weights obtained empirically from manually-annotated contraction events in a training data-set. The results show an improvement in contraction detection accuracy compared to threshold-based methods. The template-matching method is adaptable to relevant features in the input training data, and is thus less sensitive to differences in eIUP derivation or measurement variability. The method allows for improved automatic uterine contraction detection in labor EHG data, while being extensible to e.g. preterm contraction detection.