Abstract. The`canonical embedding approach' was introduced by the second author and, subsequently, it has been applied several times to prove the embeddability of certain regular extensions by groups into semidirect products by groups. In the present paper this technique is generalized so that it is suitable to handle regular extensions by inverse semigroups. As an application, B. Billhardt's embedding theorem on regular extensions of semilattices by inverse semigroups is reproved. 0. Introduction. As a possible way to generalize McAlister's P-theorem for orthodox semigroups, the second author initiated the study of embeddings of Eunitary regular semigroups into semidirect product of bands by groups. In [12] à canonical embedding approach' was developed for this purpose and was applied to prove the embeddability of E-unitary regular semigroups with regular bands of idempotents. Furthermore, this approach was used in [13] and [3], and the same ideas also appeared in [8] in connection with regular extensions of Cli ord semigroups by groups. The canonical embedding approach was generalized to the case of regular extensions of orthodox semigroups by groups in [14] and was applied to regular extensions of regular orthogroups by groups in [15]. Although the roots of the investigations of regular extensions by inverse semigroups go back to L. O'Carroll [10] and C. H. Houghton [6], the newer results on regular extensions by groups directed attention to the problem of whether similar results could be achieved for extensions by inverse semigroups. The ®rst results of this kind are due to B. Billhardt [1] and [2]. In particular, he raised the question of which regular extensions of regular orthogroups by inverse semigroups are embeddable into a !-semidirect product of a regular orthogroup by an inverse semigroup. In order to facilitate these investigations, the goal of the present paper is to develop a canonical embedding approach for regular extensions by inverse semigroups (Section 2). As an application, we reprove the main result in [1] (Section 3). Note that an alternative proof can be found in [5]. The proof presented here is an adaptation to the more general case of a proof of McAlister's P-theorem that was used several times by the second author in lectures to illustrate the canonical embedding approach (see Remark 3.6).