We present results on two related topics : 1 ) A discussion of high redshift candidates ( z > 4.5 ) and 2 ) A study of very small galaxies at intermediate redshifts , both sets being detected in the region of the northern Hubble Deep Field covered by the deep NICMOS observations at 1.6 and 1.1 microns .
The high redshift candidates are just those with redshift z > 4.5 as given in the recent catalog of Thompson , Weymann and Storrie-Lombardi , while the “ small galaxy ” sample is defined to be those objects with isophotal area \leq 0.2 ^ { \Box ^ { \prime \prime } } and with photometric redshifts 1 \leq z \leq 4.5 .

Of the 19 possible high redshift candidates listed in the Thompson et al . catalog , 11 have ( nominal ) photometric redshifts less than 5.0 .
Of these , however , only 4 are “ robust ” in the sense of yielding high redshifts when the fluxes are randomly perturbed with errors comparable to the estimated measuring error in each wave band .
For the 8 other objects with nominal photometric redshifts greater than 5.0 , one ( WFPC2 4–473 ) has a published spectroscopic redshift .
Of the remaining 7 , 4 are robust in the sense indicated above .
Two of these form a close pair ( NIC 586 and NIC 107 ) .
The redshift of the object having formally the highest redshift , at 6.56 ( NIC118 = WFPC2 4–601 ) , is problematic , since F606W and F814W flux are clearly present , and the nature of this object poses a dilemma .

Previous work by Colley et al . has suggested that compact sources in the WFPC2 HDF images are subgalactic components at redshifts z > 0.5 since they are correlated on scales less than 1 " , corresponding to physical scales of less than 8 kpc ( H _ { 0 } = 65 km s ^ { -1 } Mpc ^ { -1 } , q _ { 0 } = 0.125 ) .
We confirm these correlations in the WFPC2 data .
However , we do not detect the correlation of close pairs of galaxies on small scales in the \sim 0.65 ^ { \Box ^ { \prime } } region of the HDF that we surveyed with NICMOS .
The smaller area surveyed and lower resolution will make any real correlation more difficult to measure in these data .
We have examined averaged images of these faint ( V _ { 606 } \sim 27–29 ) , compact objects to search for extended , surrounding flux from older , fainter populations of stars .
We find no evidence from the averaged images that isolated , compact objects in the Hubble Deep Field are embedded in fainter , more extended galaxies .
For three different assumptions about possible star formation histories in these objects we set limits on the total amount of stars which could have been formed in an annulus corresponding to radii between \sim 6 to 10 kpc , which is typically a few times 10 ^ { 8 } M _ { \odot } .
We suggest that some of these objects may be protogalactic fragments .