According to World Health Organization, 5-18% births around the world are premature, and this rate is on its rise. Recent trend has been to develop computational tools which could support obstetricians in their daily practice. This work is aimed at extracting novel diagnostic features for term vs. preterm births classification based on the dynamics of contraction and non-contractions (dummy) intervals. To achieve this, 26 uterine records of 18 pregnant women, resulting in 13 term and 13 preterm deliveries, were analyzed in this study. Each record was of 30-min duration, and comprised three electrohysterographic (EHG) and one externally recorded tocographic (Toco) signal. Granger causal analysis of contraction and dummy intervals of recorded signals was performed to extract potential features. These include coupling strength and directionality indices, and were subsequently utilized by a quadratic discriminator to classify between term and preterm births. The extracted features achieved high discriminatory power of 94%, 95%, 91%, and 97% for sensitivity, specificity, classification accuracy, and area under the ROC curve, respectively. These findings show that Granger causal analysis might be adopted as a computational tool to quantify EHG-Toco dynamics for the risk-assessment of the preterm delivery.