The polarity of online news publications thematically related to COVID-19 is analysed. A collection of sentiment annotations for news articles written in the Croatian language was created and compose a new Cro-CoV-Senti-articles-2020 dataset. The news article's sentiment is derived from the reactions of portal readers. In addition, well-known sentiment analysis approaches that use lexicons and machine learning algorithms have been implemented to automatically determine the sentiment of online news. Besides, the VADER framework was used in parallel. It has been found that for the purposes of crisis communication analysis when rapid analysis solutions are needed, existing tools can be used for preliminary sentiment analysis despite some technical shortcomings. However, for a more extensive analysis of the media space and highly valuable insights, some refinements are needed. This preliminary analysis, on a sample of approximately 3,400 newspaper articles related to COVID-19, finds that readers perceive as many as two-thirds of articles negatively rather than positively.