Disclosed is a warning evaluation method for fully mechanized mining face roof disasters. The method includes: dividing fully mechanized mining face roof disaster levels into three classes of secure, less dangerous and dangerous; establishing evaluation indexes including setting load p0, resistance increase rate deltap, mining face advance speed v, roof weighting step Pe and security valve opening ratio k; acquiring values of setting load benchmark indexes p1 and p2, resistance increase rate benchmark indexes gamma1 and gamma2, roof weighting step benchmark indexes P1 and P2, mining face advance speed benchmark indexes v1 and v2 and security valve opening ratio benchmark indexes k1 and k2 by means of making statistics and analysis of mining data of a tested fully mechanized mining face; when the disaster level corresponding to any one of the indexes is dangerous, considering the fully mechanized mining face roof disaster level as dangerous. The warning evaluation method is scientific, effective, high in accuracy and convenient for production technicians to propose measures for relieving danger timely and effectively, and damage and harm caused by roof disasters are avoided.