We tested the hypothesis that fertilisation increases community and ecosystem variability while reducing predictability using annual fossil records from fertilised Lake 227, Experimental Lakes Area, Ontario, Canada. Comparison of fossil pigments from unperturbed and eutrophied periods using a median-log Levene’s test demonstrated that variability increased significantly during enrichment for total algae (chlorophyll a, sum of carotenoids), cyanobacteria (aphanizophyll, lutein-zeaxanthin), chlorophytes (pheophytin b, lutein-zeaxanthin), and cryptophytes (alloxanthin), but not for other algal taxa (chrysophytes, dinoflagellates) or herbivory (pheophorbides). Dynamic linear models (DLMs) of individual time series showed that forecast accuracy declined during enrichment for taxa which showed increased variability, while forecast uncertainty increased for all fossil pigments. DLMs of simulated data identified a strong inverse relationship between variability and predictability, suggesting that predictability will decline whenever variability increases. These findings imply that anthropogenic eutrophication of ecosystems may destabilise lakes, obscure impacts of global change, and reduce the sensitivity of whole-ecosystem experiments.