This study aimed to estimate prevalence of loneliness among older Brazilian adults over the first seven months of the COVID-19 pandemic and to identify the predictors of loneliness trajectories. Pre-pandemic data derived from face-to-face interviews of participants of the 2019-2020 Brazilian Longitudinal Study of Aging (ELSI-Brazil), which is a nationally representative study of community-dwelling individuals aged 50 years and over. Pandemic data were based on three rounds of telephone interviews among those participants, conducted from May to October 2020. Loneliness was measured by a single-item question, considering those who had at least two repeated measures. Explanatory variables included depression, living alone, leaving home in the last week, and virtual connectedness in the last month. Mixed-effects logistic regression was used to estimate odds ratios with their 95% confidence intervals (95%CI) and to investigate loneliness trajectories and their predictors. In total, 5,108 participants were included. The overall prevalence of loneliness in the pre-pandemic period was 33.1% (95%CI: 29.4-36.8), higher than the pandemic period (round 1: 23.6%, 95%CI: 20.6-26.9; round 2: 20.5%, 95%CI: 17.8-23.5; round 3: 20.6%, 95%CI: 17.1-24.6). A significant interaction (p ≤ 0.05) was evidenced only between depression and time; participants with depression showed a greater reduction in loneliness levels. Although loneliness levels in Brazil have decreased during the pandemic, this pattern is not present for all older adults. Individuals with depression had a more significant reduction, probably due to feeling closer to their social network members during the stay-at-home recommendations.