BY R. J, TILLYARD. 23 



various times and in various manners, and we are apt to lose 

 sight of the great fact of the underlying unity of descent con- 

 necting together the various groups of animals or plants upon 

 the earth. In the method proposed, no attempt will be made to 

 indicate land-area divisions or subdivisions; but the attempt at 

 subdivision or classification will be devoted to the actual contours 

 of groups. The construction of these contours is, however, a 

 matter of great difficulty. As the author is convinced of the 

 futility of attempting such a task, except under the guidance of 

 very strict and definite rules, the following scheme is here pre- 

 sented as an explanation of the method, for which the name 

 ^^ Method of Specific Contours" is proposed : — 



i. Selection of the Land- Area. 

 Any land-area, either continuous or discontinuous, may be 

 selected which may he considered to have claims to he regarded as 

 a zoogeographical unit. Without doubt the best results will be 

 obtained by the drawing of contours over the complete area of 

 one of the six main zoogeographical regions. For the study of 

 circumpolar or circumtropic distributions, it would be advisable 

 to take the total land area of the earth into discussion. Parts 

 of a region (such as Madagascar, New Zealand, or Australia with 

 Tasmania), may be studied separately with good results, provided 

 the unity of the prevailing flora or fauna of a region is not 

 destroyed by the selection of an area that has no claims to be 

 considered as a unit. 



ii. Selection of the Group. 

 The group of animals or plants selected for study by contours, 

 whether it be a single genus, group of genera, subfamily, or 

 division of higher order, must he a natural group clearly marked 

 off from its nearest allies. Genera or other groupings merely 

 based on taxonomic expediency cannot be used. For if we fail 

 to take into account any portion of a complete natural group, 

 we cannot expect to obtain a completely natural result. In 

 particular, known convergences of type must be carefully avoided; 

 but, on the other hand, when convergence is not yet proved, the 

 resulting contour may give valuable evidence as to its existence, 



