Given a level set E of an arbitrary multiplicative function f , we establish, by building on the fundamental work of Frantzikinakis and Host [15,14], a structure theorem which gives a decomposition of 1 E into an almost periodic and a pseudo-random part. Using this structure theorem together with the technique developed by the authors in [3], we obtain the following result pertaining to polynomial multiple recurrence.Theorem. Let E = {n 1 < n 2 < . . .} be a level set of an arbitrary multiplicative function with positive density. Then the following are equivalent:• E is divisible, i.e. the upper density of the set E ∩ uN is positive for all u ∈ N;• E is an averaging set of polynomial multiple recurrence, i.e. for all measure preserving systems (X, B, µ, T ), all A ∈ B with µ(A) > 0, all ℓ 1 and all polynomialsWe also show that if a level set E of a multiplicative function has positive upper density, then any self-shift E − r, r ∈ E, is a set of averaging polynomial multiple recurrence. This in turn leads to the following refinement of the polynomial Szemerédi theorem (cf. [4]).Theorem. Let E be a level set of an arbitrary multiplicative function, suppose E has positive upper density and let r ∈ E. Then for any set D ⊂ N with positive upper density and any polynomials p i ∈ Q[t], i = 1, . . . , ℓ, which satisfy p i (Z) ⊂ Z and p i (0) = 0 for all i ∈ {1, . . . , ℓ}, there exists β > 0 such that the set n ∈ E − r : d D ∩ (D − p 1 (n)) ∩ . . . ∩ (D − p ℓ (n)) > β has positive lower density.