This paper proposes a new approach based on time-varying copulas to test for the presence of increases in stock market interdependence (also known as shift contagion) after a financial crisis. We discuss the importance of considering simultaneously separate breaks in volatility and dependence. Without such consideration, the contagion test turns out to be biased. A sequential algorithm is proposed to tackle this problem. Applied to the recent 1997 Asian crisis, the analysis confirms that breaks in variances always precede those in the dependence parameter. Moreover, a significant 'J-shape' evolution of the dependence parameter is detected, supporting the idea of shift contagion.