We address the issue of panel cointegration testing in dependent panels, showing by simulations that tests based on the stationary bootstrap deliver good size and power performances even with small time and cross-section sample sizes and allowing for a break at a known date. They can thus be an empirically important alternative to asymptotic methods based on the estimation of common factors. Potential extensions include test for cointegration allowing for a break in the cointegrating coefficients at an unknown date.