The invention discloses a role mining optimization modeling method meeting a user role-usage cardinality constraint. It is assumed that q candidate roles R1, R2, ..., Rq exist, and a set of all the candidate roles also can be represented as R by a matrix; meanwhile, a new Boolean marking variable set {d1, d2, ..., dq} is defined, and when dj is equal to 1, the role j appears; otherwise, the role jdoes not appear; consequently, a formula defined in the specification is workable; when dj is equal to 1, at least one value {c1j, c2j, ..., cnj} is 1, namely, at least one user has the role j, wherein Vj:dj is greater than or equal to dij, and i is greater than or equal to 1 and less than or equal to n; by loosening an integrity constraint in a first problem, a better role result can be mined probably, which is an original intention of research on a second problem; and for modeling the problem, a slack variable x'it is introduced, and under the condition of not exceeding a threshold delta, matrix reconstruction is acceptable.