International tourist arrivals increased by over 4,000% during the last 60 years, and as a labour-intensive business, tourism destinations and suppliers strongly depend on precise predictions of tourism demand. This study compares an autoregressive approach to predict tourism demand which is using past arrivals as input with an approach which predicts arrivals based on big data information as additional input, like the destination price level or the web search traffic per sending country, respectively. As prediction methods, the study uses the statistical approach of the linear regression and the data mining technique k-nearest neighbour (k-NN). Both approaches are executed and evaluated for the leading Swedish mountain destination Åre on the base of arrival data and big data sources for the time period 2005–2012. Study results show that (1) big data information sources can significantly increase the prediction performance of tourist arrivals compared to using past arrivals alone (i.e. autoregressive approach) and (2) data mining techniques (i.e. k-NN) can outperform statistical approaches, like linear regression.