Many remote sensing products that are useful for time series analysis and seasonal monitoring studies are offered in form of composites. A composite combines a number of observations of a defined period and selects or computes one value. This results in observations sampled at varying time intervals that rules out a high number of time series analysis techniques. This study investigates the impact of either using the actual day of observation to generate a time series from composites or assuming the starting or middle day of the compositing period. For this study 16-day MODIS VI composites of 1km spatial resolution from Terra and Aqua were employed. A 1100x500km region in central Mexico served as study site. Statistical measures including temporal cross-correlation and the root mean square error were used for time series analysis. A temporal shift of approximately seven days with a high variability is introduced when using the starting day of the compositing period. The middle day mitigates the mean error close to zero but still shows a high error variability. Only time series that take into account the day of observation and estimate from that samples at equidistant intervals can be used for a correct estimation of temporal characteristics.