This paper provides an extensive Monte-Carlo comparison of severalcontemporary cointegration tests. Apart from the familiar Gaussian basedtests of Johansen, we also consider tests based on non-Gaussianquasi-likelihoods. Moreover, we compare the performance of these parametrictests with tests that estimate the score function from the data using eitherkernel estimation or semi-nonparametric density approximations. Thecomparison is completed with a fully nonparametric cointegration test. Insmall samples, the overall performance of the semi-nonparametric approachappears best in terms of size and power. The main cost of thesemi-nonparametric approach is the increased computation time. In largesamples and for heavily skewed or multimodal distributions, the kernel basedadaptive method dominates. For near-Gaussian distributions, however, thesemi-nonparametric approach is preferable again.