Russian Longitudinal Monitoring Survey (RLMS) is one of the main data sources for econometrical analysis of unemployment in Russia, along with the data from Public Employment Offices (PEO). RLMS data contain the information about non-registered unemployment which is not provided by PEO. Another advantage of RLMS data is that they are accessible for wide range of researchers. However, serious problems appear in analysis. Survey takes place only once a year, which makes measuring the duration of unemployment spells difficult and also leads to selectivity bias. The bias appears because longer spells are more likely to be observed while short spells that take place between the rounds of the survey remain unnoticed. In this article, a method for duration analysis using RLMS data is proposed and applied to the data concerning the duration of unemployment in Russia in 2000–2005.