In current research, social relationships are increasingly recognized for their positive associations with cognitive outcomes in older adults. One of the most vulnerable groups for cognitive decline are very old adults (80+ years). However, they are relatively underrepresented in the field. Therefore, our study aimed to test if social relationships remain a protective factor against cognitive decline in very old age, using a representative sample from the Study of Quality of Life and Well-Being in North-Rhine Westphalia (NRW80+ Study). We hypothesized that social characteristics would be positively associated with global cognition and episodic memory cross-sectionally and would predict cognitive performance two years later. 1.207 very old adults were included in the representative, cross-sectional analyses, and 639 in the panel analyses. They were aged between 80 and 103 years and showed no signs of dementia. The associations between various social aspects and cognitive functions were investigated using hierarchical linear regression, controlling for relevant sociodemographic and health characteristics. Cross-sectionally, leisure engagement was positively associated with episodic memory (β = 0.53 [0.26, 0.79], p < .01) and global cognition (β = 0.50 [0.22, 0.79], p < .01), while overall network size was positively associated with global cognition (β = 0.04 [0.02, 0.07], p < .01). In contrast, we observed no associations between baseline social relationships and cognitive functions two years later. The findings suggest that while social relationships are associated with cognitive functions in very old age, short-term protective effects such as over two years may be less robust compared to other age groups.