Studying end of life in dementia patients is challenging because of ill-defined prognoses and frequent inability to self-report. We aim to quantify and compare (1) feasibility and (2) sampling issues between prospective and retrospective data collection specific to end-of-life research in long-term care settings. The observational Dutch End of Life in Dementia study (DEOLD; 2007 to 2011) used both prospective data collection (28 facilities; 17 nursing home organizations/physician teams; questionnaires between January 2007 and July 2010, survival until July 2011) and retrospective data collection (exclusively after death; 6 facilities; 2 teams, questionnaires between November 2007 and March 2010). Prospective collection extended from the time of admission to the time after death or conclusion of the study. Prospectively, we recruited 372 families: 218 residents died (59%) and 184 (49%) had complete physician and family after-death assessments. Retrospectively, 119 decedents were enrolled, with 64 (54%) complete assessments. Cumulative data collection over all homes lasted 80 and 8 years, respectively. Per complete after-death assessments in a year, the prospective data collection involved 37.9 beds, whereas this was 7.9 for the retrospective data collection. Although age at death, sex, and survival curves were similar, prospectively, decedents' length of stay was shorter (10.3 vs. 31.4 mo), and fewer residents had advanced dementia (39% vs. 54%). Regarding feasibility, we conclude that prospective data collection is many fold more intensive and complex per complete after-death assessment. Regarding sampling, if not all are followed until death, it results in right censoring and in different, nonrepresentative samples of decedents compared with retrospective data collection. Future work may adjust or stratify for dementia severity and length of stay as key issues to promote comparability between studies.