In rapidly urbanising areas, where maps are seldom up-to-date, recent aerial 
photography, with its wealth of detail on streets and houses, can be and often 
is a highly valued source in the planning and execution of such a census. The 
photographs make it possible to delineate enumeration districts for which the 
field data will be aggregated, to distribute these areas over the hundreds of 
officials who will collect the data in the field, and to organise procedures for 
checking the field workers by team leaders. Normally, a census is carried out 
for a whole country in one year, which means that proper planning of photo- 
graphy has to be included in the planning of such a census. For rapidly growing 
metropolitan areas, recent photography is more important than for remote 
villages; cities should thus be flown at the latest time possible. 
The main problems with censuses are the very high costs involved and 
the difficulty of controlling the large ad hoc organisation that has to be set up. 
In countries where a full census is therefore not considered feasible, a census 
based on sample surveys can at least fulfill partially the need for information. 
Such a census has recently been carried out in Afghanistan (Dayal and Khair- 
zada 1976). In a sample census, the aerial photograph plays a triple role. It 
provides not only the topographical base material, but serves also as a source 
of information for stratified sampling according to types of residence, and 
makes possible the extension from sample to universum because it permits 
measuring the size of both sample area and total area, or counting housing 
units in both areas. 
Such an extension is based on assumptions, which are, however, not yet 
properly tested. Many questions have still to be answered by research. What is 
known of the correlation between a classification of residential areas based on 
artefactual information from the aerial photograph, and a classification of 
residential areas based on social and economic data from census or registration? 
Moreover, how homogeneous is each class in terms of spatial distribution? 
This last question concerns the areal bias: does it make a difference whether 
a sample area is chosen in one part of a residential area or in another part? 
And then, how sensitive to areal bias are the different characteristics of the 
population? It may be assumed, for example, that estimates of the total pop- 
ulation derived from sample surveys are better than estimates of the number 
of people not locally born (migration). Analyses in terms of vectors (distance 
to city centres, or better perhaps, if sequential photography is available, date