Background: Antifreeze proteins(AFPs), also known as thermal hysteresis proteins, are ice-binding proteins.AFPs can adsorb to ice crystal surface and inhibit the growth of ice crystals in solution.So in AFPs solution the temperature of growth point of ice crystal is below the melting point temperature of ice crystal.But the interaction between AFPs and ice crystal is not known completely.Analyzing physicochemical properties of AFPs sequences is very significant to understand the ice-protein interactions.And using these important physicochemical properties, we can further accurately recognize AFPs.Methods: Two datasets were constructed.The first dataset included 597 non-redundant AFP sequences in the UniProtKB database.The second dataset included 5124 non-redundant non-AFP sequences from the PISCES server.We calculated the amino acid and dipeptide composition of 597 AFPs.According to the hydrophobicity, acid-base property, side-chain mass and polarizability of the amino acids, the amino acids were respectively divided into 6 groups,3 groups, 3 groups,4 groups.And the pseudo amino acid composition for every physicochemical property was calculated.We used Support Vector Machines (SVM) to predict AFPs based on Physical and chemical characteristics which were obtained by statistical analysis.Results: Through the analysis of AFPs amino acid sequence, we find that Alanine and threonine account for a large proportion of the antifreeze protein amino acid composition.Incorporating amino acids composition and the physicochemical properties distribution into the input vector by SVM, high accuracy with good sensitivity was achieved in the jackknife test.Conclusions: Strongly hydrophobic property plays an important role in the antifreeze protein and ice interactions.And hydrophobic amino acid groups are main part of ice-binding region on the surface of AFPs.Analyzing physicochemical properties of AFPs sequences is very significant to understand the ice-protein interactions .