Many terrestrial biomes are experiencing intensifying human land use. However, reductions in the intensity of agricultural land use are also common and can lead to agricultural land abandonment. Agricultural land abandonment has strong environmental and socio-economic consequences, but fine-scale and spatially explicit data on agricultural land abandonment are sparse, particularly in developing countries and countries with transition economies, such as the post-Soviet countries of Eastern Europe. Remote sensing can potentially fill this gap, but the satellite-based detection of fallow fields and shrub encroachment is difficult and requires the collection of multiple images during the growing season. The availability of such multi-seasonal cloud-free image dates is often limited. The goal of our study was to determine how much "missing" Landsat TM/ETM. + images at key times in the growing season affect the accuracy of agricultural land abandonment classification. We selected a study area in temperate Eastern Europe where post-socialist agricultural land abandonment had become widespread and analyzed six near-anniversary cloud-free Landsat images from "Spring", "Summer" and "Fall" agriculturally defined seasons for a pre-abandonment-time I (1989) and post-abandonment-time II (1999/2000). Using a factorial experiment, we tested how the classification accuracy and spatial patterns of classified abandonment changed over all possible 49 image-date combinations when mapping both "abandoned arable land" and "abandoned managed grassland". The conditional Kappa of our best overall classification with support vector machines (SVM) was 90% for "abandoned arable land" and 72% for "abandoned managed grassland" when all six images were used for the classification. Classifications with fewer image dates resulted in a substantial decrease of the conditional Kappa (from 93 to 54% for "abandoned arable land" and from to 75 to 50% for "abandoned managed grassland"). We also observed substantial decrease in accurate detection of land abandonment patterns when we compared our best overall classification with the other 48 image date combinations (the Fuzzy Kappa, a measure of spatial similarity, ranged from 25.8 to 76.3% for "abandoned arable land" and from 30.4 to 79.5% for "abandoned managed grassland"). While the accuracy of the different abandonment classes was most sensitive to the number of image dates used for the classification, the seasons captured also mattered, and the importance of specific seasonal image dates varied between the pre- and post-abandonment dates. For "abandoned arable land" it was important to have at least one "Spring" or "Summer" image for pre-abandonment and as many images as possible for post-abandonment, with a "Spring" image again being most important. For "abandoned managed grassland" no specific seasonal image dates yielded statistically significantly more accurate classifications. The factor that influenced the accurate detection of "abandoned managed grassland" was the number of multi-seasonal image dates (the more the better), rather than their exact dates. We also tested whether SVM performed better than the maximum likelihood classifier. SVM outperformed the maximum likelihood classifier only for "abandoned arable land" and only in image-date-rich cases. Our results showed that limited image-date availability in the Landsat record placed substantial limits on the accuracy of agricultural abandonment classifications and accurately detected agricultural land abandonment patterns. Thus, we warn to use agricultural land abandonment maps produced with the sub-optimal image dates with caution, especially when the accurate rates and the patterns of agricultural land abandonment are crucial (e.g., for LULCC models). The abundance of agricultural abandonment in many parts of the world and its strong ecological and socio-economic consequences suggest that better monitoring of abandonment is necessary, and our results illustrated the image dates that were most important to accomplishing this task.