Russian Gapping: evidence for deletion ATB-movement approaches to Gapping (Zoerner 1995, Steedman 2000, Johnson to appear) explain why Gapping is restricted to coordinate sentences and cannot affect an embedded verb when the matrix one is retained. At the same time, all such approaches face the linearization problem in SVO languages. Deletion approaches to Gapping (Jayaseelan 1990 and others) mirror the advantages and disadvantages of the ATB approaches. Russian has a special possibility for Gapping which undermines the ATB approach. That is Gapping in sentences like (1), where predicate coordination precedes the gapped sentence, and the antecedent verb is in the last conjunct of that coordinate structure. If the structure of (1) is roughly as in (2), with the first two predicate forming a coordinate constituent which excludes the third predicate, then any ATB movement of a material contained in the gapped and antecedent clause will violate the Coordinate Structure Constraint when the antecedent verbs will have to cross &P2. It can be argued that (2) is correct at least when the sentences which precede Gapping have identical subject, as in (1). There is evidence that when i "and" conjoins same subject predicates with the subject expressed before the first predicate, the correct structure is (3) rather than (4) (I accept without argument that subject ends up in AgrSP and the verb ends up in TP in Russian; the projections specification is not relevant for the present argument): (3) [AgrSP DP [TP [TP1] and [TP2]]] (4) [[AgrSP DP1 TP1] and [AgrSP DP1 TP2]] Three kinds of evidence are adduced in favor of (3). First, it is argued that (3), but not (4) accounts for impossibility of agreement clashes under numeral subjects, with plural agreement in one conjunct and singular neutral agreement in the other one (5). Second, (3), but not (4) explains why in ATB Wh-questions, Wh-subject only allows the ATB-reading, whereas Wh-object and other TP-internal Wh-phrases allow both ATB and paired readings (6) vs.(7)). Third, it is shown that (3) correctly predicts that when a coordinate structure is embedded under the complemenizer cto, the subject of the second conjunct is obligatorily understood as identical to the subject of the first conjunct, despite of the possibility for the matrix subject to bind a zero subject under cto outside coordination ((8), (9)). It is shown that if the position hosting the subject (presumably the Spec of the AgrSP) existed in the second conjunct of (8), there would be nothing what could block binding of a zero element in that position by the matrix subject. If the structure in (3) is correct, then (1) can be accounted for only if we accept a deletion rather than ATB approach to Gapping in Russian. Further evidence against ATB analysis comes from impossibility of wide scope negation in Russian Gapping. The wide scope in English Mr Smith cannot dance and Mrs Smith sing is accounted for via ATB movement of the negation, see Johnson, to appear. If ATB were possible in Russian Gapping, the wide scope would also be expected in Russian, contrary to fact. The alternative, deletion approach to Gapping suggests that some phrase containing V, e.g. TP, is deleted after the expressed material is evacuated out of it. This is supported by ungrammaticality of the mirror image of (1), where the gapped verb is in the first conjunct of a coordinate structure with i "and"(10). If the structure of coordination with i is as in (3), then the expressed material cannot be evacuated from TP1 due to the Coordinate Structure Constraint, yielding (10) ungrammatical under the deletion approach. The paper is concluded by some typological parallels, suggesting that both of the two competing approaches can be correct: some languages have ATB Gapping, like English, and others, like Russian, have deletion Gapping. The clusters of characteristics essential for each type of Gapping are listed.