An improved concept of the best analogs method is used to reconstruct the climate of the last glacial maximum from pollen data in Europe. In order to deal with the lack of perfect analogs of fossil assemblages and therefore to obtain a more accurate climate reconstruction, we used a combination of pollen types grouped according to plant phenology and present climate constraints rather than pollen percentages for each individual taxon. The distribution of pollen taxa into plant functional types (PFTs) is aimed to reflect the vegetation in terms of biomes which have a wider distribution than a species. The climatic variables are then calibrated on these PFTs using an artificial neural network technique. The use of PFTs allowed us to deal with situations where pollen assemblages have only partial modern analogs. The method is applied to the glacial steppic vegetation in Europe, using 15 pollen records. North of the Pyrenees–Alps line, the reconstructed temperatures were lower than today: −30 ± 10°C for the temperature of the coldest month ( T
c) and −12 ± 3°C for the annual mean ( T
ann). South of that line, T
c and T
ann anomalies were respectively, −15 ± 5°C and −10 ± 5°C. The available moisture index and annual precipitation were also lower than present: −60 ± 20% north of Mediterranean Sea, (−800 ± 100 mm for precipitation). In Italy and Greece, the available moisture was 20% lower, with a precipitation anomaly of ca. −600 ± 200 mm. Southward, the moisture index was close to that at present (±20%), and precipitation was lower (−300 ± 300 mm).