SCIENTIFIC ABSTRACT We propose deep imaging with ISOCAM of fields containing gravitationally lensed galaxies (giant arcs) with redshifts up to 2.2. In this way we take advantage of the vast ``gravitational telescopes'' constituted by distant galaxy clusters to study the infrared flux from the lensed galaxies themselves and to probe these fields to high redshift for other, IR luminous, lensed objects. These observations will: a) - generate infrared images of known giant arcs. b) - provide the infrared flux of the giant arcs and hence extend our knowledge of the properties of field galaxies at high redshift. c) - search for infrared arcs that would be produced by a primeval population of infrared luminous galaxies. d) - will be complementary to ISOCAM proposals to make deep integrations on galaxy clusters. OBSERVATION SUMMARY Given the R magnitudes for these arcs, and assuming normal galaxy colours, we can calculate their flux in the M band (4.8 microns). This is, for the arcs to be studied here, in the range 3 to 12 microJanskys per square arcsecond. We will use CAM in micro--scanning mode (5x5 microscan, 1.3 pixel step size) in the following configurations, for all targets: - filter LW1 (4 to 5 microns), on chip integration time (tint) 20 seconds, pixel f.o.v. 3 arcseconds, exposure time 3600 seconds and - filter LW6 (7 to 8.5 microns), on chip integration time (tint) 20 seconds, pixel f.o.v. 3 arcseconds, exposure time 3600 seconds. These configurations will allow us to reach S/N ratios of 3 and 11.5 respectiv- ely, in the two configurations described above, for arcs having fluxes of 8 microJanskys per square arcsecond. Therfore, while this is a difficult measure- ment, it is nonetheless well within the bounds of possibility that the arcs can be detected with CAM given its sensitivity. Quite apart from the sensitivity of CAM, another potentially serious constraint is the possibility of confusion of the IR arc images with foreground galaxies in the lensing cluster. The diameter of the ISO diffraction spot is roughly, in arcseconds, 0.84 times the wavelength. This implies diffraction spot sizes of 3.8 and 6.5 arcsedonds, in the filters chosen here. We have therfore chosen the 3 arcsecond per pixel p.f.o.v. with the LW1 filter and the 3 arsecond p.f.o.v. with the LW6 filter. These are a compromise between sampling of the PSF for the wavelenghts employed and achievable S/N and should avoid serious confusion with nearby sources for the arcs studied here except for the case of A370, where careful modelling of the foreground cluster will be necessary in order to compensate, during data reduction, for source confusion. Observing between 4 and 5 microns reduces flat-fielding problems inherent at longer wavelengths where the background is higher. The choice of the LW1 filter and the 3 arcsec per pixel field of view maximises the resolution and the sampling of the PSF for useful S/N, vital if the confusion limit is to be avoided. The combination of the 3 arcsecond p.f.o.v. and the LW6 filter allows us to take advantage of any growth in the IR flux going further into the IR (if the lensed galaxies are AGNs) and provides greater sensitivity due to the larger pixel size. For the 3" p.f.o.v, the Zodiacal background flux is much greater than the source flux. Then detector responsive transients which can occur when a source moves over the array, should not be too severe, since the array sees essentially uniform illumination, even during microscanning, after the initial transient at source acquisition. The transient behaviour of the detector should be better for the higher pixel signal encountered with the higher background seen in the LW6 filter. Micro-scanning reduces the impact of residual flat-fielding noise and enhances the possibilities for later application of super-resolution techniques.