We compared four methods for combining separate fragmentary phenological time series into a single long reliable series. The systematic linear effect of differences in observers, genotypes, geography and climate at the observation points produces disturbing variation in the observations and bias in the means of some time points. The three methods based on the adjustment of individual series eliminated the disturbing variation and bias. The methods were compared based on phenological observations of bud burst in birch (Betula pendula Roth). The method based on a linear mixed model of analysis of variance and the maximum likelihood estimation was considered preferable to the other methods.