This study aimed (i) to assess the relative weight of different predicting variables, related to site and grassland characteristics, in determining the fatty acid (FA) composition of permanent grasslands and (ii) to verify if and at which extent the same variables have a significant predicting role for the FA composition of the derived bulk cow milk. Available data collected from 2003 to 2016 by three European research institutions were used. A dataset (n = 144) was built up including: (i) the proximate and FA compositions of herbage, (ii) site altitude and climatic conditions, (iii) grasslands botanical composition (main family groups), herbage phenology and growing cycle, (iv) herd characteristics and (v) milk gross and FA compositions. Prediction models for herbage FA profile were highly reliable (R2 adjusted ≥ 0.80) for C18:3 n3, total monounsaturated FA and total polyunsaturated FA (Σ PUFA). Reliable predictions (R2 adjusted ≥ 0.60) were obtained for the other main herbage FA. The relative predicting weight was higher for phenology and proximate composition than for botanical composition. Among proximate composition variables, dry matter showed the highest relative weight in determining the majority of the predicting models. Prediction models for milk FA profile were reliable (R2 adjusted ≥ 0.60) for Σ de novo FA, C16:0, C18:3 n3, total even-chain saturated FA (Σ ECFA), Σ PUFA and the C18:1 c9/C16:0 ratio, while they were moderately reliable (R2 adjusted ≥ 0.50) for C18:1 t10+t11, C18:2 n6, C18:2 c9t11 and the C18:2 n6/C18:3 n3 ratio. Diet composition, site altitude and grazed grassland features concurred to predict milk FA. Fresh herbage proportion in the diet played a relevant role in determining the milk proportion of several FA. Phenology was the main driver for the C18:2 c9t11 and Σ PUFA proportions, negatively affecting them. Herbage phenology fitted milk FA better than proximate and botanical compositions.