In this paper two estimation and testing procedures for cointegration are compared for variables with deterministic trends. The first procedure consists of static regression estimator and a residual unit root test; the second estimator and (Wald) test statistic are derived from a dynamic regression. The asymptotic properties are derived, and critical values for the Wald test are simulated. A Monte Carlo simulation study suggests that in small samples the performance of the two procedures critically depends on the parameters of the data generating process.