SCIENTIFIC ABSTRACT: We propose to perform deep integrations on a selection of `empty fields' at high galactic latitudes, using the most sensitive ISOCAM filters. The two target fields (a Southern field at RA=03:13 DEC=-55:25 and the Lockman Hole at RA=10:49 DEC=+57:37) are characterized by the lowest possible cirrus emission and avoid bright sources: they have been selected on the basis of existing deep surveys in the optical, near-IR, radio and X-ray bands. One of the major scientific aims of this program is to investigate the evolution of the infrared properties of galaxies with redshift. The evolutionary properties of starburst galaxies emitting most of their energy in the infrared is one important question raised after IRAS survey, which ISO can resolve. Galaxies at high redshift might be detectable in the ISOCAM bands because of the strong and positive K correction on the stellar emission of these galaxies in the infrared. We also expect to detect statistically significant numbers of Active Galactic Nuclei (AGNs) at high redshifts (z >1), so that their cosmological evolution can be tested. Primeval galaxies at moderate redshifts may also be found. Complementary observations of the same field with ISOPHOT will be performed to study the bulk of the dust emission in these galaxies. The analysis of the deflection distribution P(D) below the confusion limit will permit the extension of source counts to significantly fainter fluxes and enable one to study the structure of the diffuse background on scales intermediate between those of COBE and of ground-based K band surveys. Spectral information, including spatial correlations between images of a field in different wavebands, will be of great importance in achieving these goals. In addition, this program offers the possibility of finding brown dwarfs in the solar neighbourhood. ISOCAM simulated performances appear perfectly suited to the deep integrations we propose. OBSERVATION SUMMARY: Two fields are mapped with ISOCAM LW channel with 6 arcsecond pixels. The Southern one (the `Marano field': MF) and the Northern one (the `Lockman hole': LH). For each field, we perform a shallow survey (SS) at moderate sensitivity and with a large area (about 1 square degree) and a deep survey (DS) at a larger sensitivity and with a smaller area (0.17 and 0.11 sq. deg. for MF and LH resp.). The DS is done with 2 filters: LW2 and LW3, the SS is done with LW3, in micro-scanning concatenated AOT mode (CAM01). The same pixel of the sky is seen by different parts of the camera during the survey in order to detect any systematic effects. If a sub-field is crowded this technique of concatenated micro-scanning AOT could allow the determination of an accurate flat-field when reducing the data. Otherwise, reduction of data with beam-switching technique should be enough. Integration times are 10s for LW2 and 5s for LW3. Total observing times (including overheads) are (in hours) MFDS_LW2 MFDS_LW3 MFSS_LW3 MFSS_PHT LHDS_LW2 LHDS_LW3 LHSS_LW3 30 20 20 20 20 13 14 Expected final noise figures are (for 4 sigma) DS_LW2: 0.034mJy, DS_LW3: 0.13mJy, SS_LW3: 0.38mJy. Confusion by cirrus clouds and galaxies should be negligible. A long wavelength map of the Southern SS with ISOPHOT C100 at 90 microns in a raster mode (PHT22) is also covered with the 3 by 3 array of 43.5 arcsecond pixels. A sensitivity of (4 sigma) 2.8mJy is reached in about 100 seconds. It is slightly below the expected confusion due to galaxies and well above the confusion due to cirrus clouds (because the region is selected in a dark IRAS area). CONCATENATION: As this survey can be completed only by a large total integration time, we have adopted a strategy of adding up several two-hour (maximum) long concatenated AOTs which cover a scan or two at constant declination.