Abstract The objective of this study was to compare the safety and efficacy of atosiban and ritodrine in the treatment of threatened preterm labour (TPL) and to analyse the predictive factors of preterm delivery. We retrospectively sampled data on 380 women hospitalised for TPL (24-35 weeks' gestation), in our clinic between 2004 and 2007. All were subjected to tocolysis with ritodrine and/or atosiban. Data were analysed using R (version 2.12.1), considering p < 0.05 as significant. We had 69 women treated with atosiban, 242 treated with ritodrine and 69 treated with ritodrine changed for atosiban, if adverse effects occurred. In the multivariate logistic regression, the use of atosiban vs ritodrine does not play any role in delaying delivery after 48 h or 7 days, whereas the cervical change at the digital examination, high contractions pre/post-therapy ratio, pPROM, cervical length and fibronectin result as predictive factors for both delivery before 48 h or 7 days. Maternal adverse drug effects were significantly more frequent in patients treated with ritodrine, and one single case of pulmonary oedema was observed. We found fewer side-effects in the atosiban than in the ritodrine group and no difference in efficacy. Moreover, the most predictive factors for preterm delivery were fibronectin test, pPROM, digital vaginal examination and uterine contraction persistence. We believe that predictive capacity of these tests could give the opportunity for targeting therapy and limiting drug side-effects and cost.