Initially overlooked by investors, the sovereign credit risk has been reassessed upwards since the 2000's which has contributed to awaken the interest of speculators in sovereign CDS. The growing need of accurate forecasting models has led us to fill the gap in the literature by studying the predictability of sovereign CDS volatility, using both linear and non-linear GARCH-class models. This paper uses data from 38 worldwide countries, ranging from January 2006 to March 2017. Results show that the CDS markets are subject to periods of volatility clustering, nonlinearity, asym-metric leverage effects and long-memory behavior. Using 7 heteroskedastic and no heteroskedastic-robust statistic criteria, results show that the fractionally-integrated models outperform the basic GARCH-class models in terms of forecasting ability and that allowing flexibility regarding the persistence degree of variance shocks significantly improves the model's suitability to data. Despite the divergence in the economic status and geographical positions of the countries composing our sample, the FIGARCH and FIEGARCH models are mainly found to be the most accurate models in predicting credit market volatility. JEL Classification: G15, G17, C58.