Attitudes towards the risk of damage by biological control agents to plants other than the target weed, and the use of host-specificity testing to minimise that risk, have changed through the history of biological control of weeds.

In the early programs by Hawaii, no host-testing was undertaken (Waterhouse and Norris 1987). Observations were made in the field in Mexico, and if the insects were seen feeding only on lantana, it was assumed that they were sufficiently host-restricted for safe introduction. Hawaii continued to rely on field observations in the country of origin to determine host-specificity, until at least the 1950s. The unique and restricted flora of Hawaii, with few relationships to plants in other continents, probably reduced any problem of attack on plants other than the target weed. However, problems did arise when insects were introduced into other countries on the basis of their use in Hawaii and without further tests. For example, the lantana tingid Teleonemia scrupulosa was introduced into East Africa in the 1960s without further testing, where it caused problems by moving onto the crop sesame when the lantana was defoliated. There was also attack on teak trees. Economic damage was not great, but the attack caused problems and it is likely that the insect would not have been introduced if host-testing had been carried out first.

Host-specificity testing was first used in Australia in the 1920s, in the major campaign against the prickly pears (Dodd 1940). Field observations, where insects were seen to be feeding only on Cactaceae in the wild, together with the known association of the insect type with Cactaceae, were still seen as the chief proof of host specificity. Host-testing was undertaken mainly to satisfy the general public that crop plants would not be damaged. Initially, both choice tests and no-choice tests were used, but the choice tests were quickly abandoned in favour of no-choice tests. In choice tests, candidate agents are confined for varying periods of time with the test plant species together with the target weed. In no-choice tests, candidate agents are confined usually with only one test plant species and without the target weed. The tests used were starvation tests, that is, the insects were confined on the test plant until they died or developed through to the next stage. As the purpose of the tests was to prove that crop plants were safe, only plants of economic importance were tested; there was no concern for native Australian plants other than their economic value.

The conservative principle was adopted from the start (Dodd 1940): that is, an insect was rejected if it could complete development on a test plant, even if oviposition on the plant would not normally occur, and even where the insect was known not to attack the test plant in the field in its native range. Similarly, if one species in a genus was considered unsafe, the whole genus was rejected. Several stages were tested; always newly-hatched larvae or nymphs, and usually half-grown larvae and adult oviposition as well.

By the 1950s, the next major period of activity, host-testing was seen as an essential part of a weed biological control program. No-choice or sequential host-specificity tests of varying duration became the accepted methodology for the determination of host specificity until the late 1960s (Harris and Zwolfer 1968). Sequential tests involve the sequential presentation of a series of test plants in a no-choice situation, with each plant species usually exposed to the candidate agent for a relatively short time.

In the late 1960s Harley (1969) advocated choice as opposed to no-choice tests. It was argued that choice tests are a more 'natural test' of host range, as the target weed will be usually be present in mixed stands with other plants in the field situation (Cullen 1990). Choice tests may lead to fewer incidences of feeding on test plant species and less rejection of 'safe' insects. Discussions on the 'best' methods for host-specificity testing continue, and a mixture of choice and no-choice tests is usually used.