The Hubble Deep Field: Pointing at Nothing, Finding Everything

A calculated gamble with the director's own time

The Hubble Deep Field was not a scheduled program that survived peer review — it was a bet. Robert Williams, then Director of the Space Telescope Science Institute, held a personal allocation called Director's Discretionary Time, and in 1995 he chose to spend a large chunk of it — around 150 orbits of the telescope, ten precious days — on a single blank field. Many colleagues thought it reckless. There was no guaranteed target, no bright object to anchor the shot; if the sky really was empty there, Hubble would return ten days of noise and Williams would have squandered a resource worth millions of dollars.

The reasoning behind the gamble was sound. Hubble sits above the atmosphere, so it suffers none of the atmospheric seeing that smears ground-based images and none of the sky brightness from airglow. Its resolution is limited only by diffraction — about 0.05 arcseconds at visible wavelengths — and its background is genuinely black. On such a telescope, faintness is limited only by how long you are willing to collect photons. Stack enough exposures and the sky keeps getting deeper, revealing objects that are individually far too faint to register in any single frame.

Choosing a patch of pure nothing

Picking the field was an exercise in deliberate emptiness. The team wanted a spot with no bright stars (which would bleed and saturate the detector), no nearby galaxies, no known radio, X-ray, or infrared sources, and as little foreground contamination as possible. That meant observing away from the plane of the Milky Way, at high galactic latitude, so the line of sight punches out of our galaxy's disk of stars, gas, and dust rather than through it — minimizing interstellar extinction and the clutter of foreground stars.

• The chosen field lies in Ursa Major, just above the handle of the Big Dipper, at approximately RA 12ʰ 36ᵐ 49ˢ, Dec +62° 13′.
• Its high northern declination placed it in Hubble's continuous viewing zone, so the target never set behind the Earth as the telescope circled the planet every 96 minutes — maximizing usable observing time each orbit.
• The field spans only about 2.6 arcminutes on a side, a solid angle of roughly 5.3 square arcminutes — about one twenty-four-millionth of the full 41,253-square-degree sky.

In short, the astronomers picked the most boring-looking piece of sky they could find, on purpose. What made it scientifically valuable was precisely that it looked like nothing.

How the image was built: 342 exposures, four colors

Between 18 and 28 December 1995, Hubble's Wide Field and Planetary Camera 2 (WFPC2) took 342 separate exposures of the field. They were split across four filters spanning the near-ultraviolet to the near-infrared: F300W (~300 nm, UV), F450W (~450 nm, blue), F606W (~606 nm, a wide visual band), and F814W (~814 nm, near-infrared I). Total integration exceeded 100 hours, with the deepest single filters accumulating well over a day of exposure each.

Individual frames are dominated by noise. The magic is in the stacking. Real sources add up coherently while random detector read-noise, sky signal, and cosmic-ray hits — which pepper every space-based CCD frame — average down or are rejected. The team used a technique called dithering: the telescope was nudged by fractions of a pixel between exposures, then the frames were recombined (a process later formalized as "drizzling") to reconstruct detail finer than the native pixel grid and to fill gaps between the four CCD chips. The result reached about 30th magnitude — objects delivering only a handful of photons over the entire campaign. On the standard logarithmic magnitude scale, 30th magnitude is roughly 4 billion times fainter than the faintest star the unaided eye can see (~6th magnitude).

What the darkness turned out to contain

When the final composite was assembled and released in January 1996, the empty patch was revealed to hold about 3,000 galaxies in an area smaller than a dime held 23 metres away. Almost every smudge of light in the frame is an entire galaxy of billions of stars; only a tiny handful of the point-like objects are foreground Milky Way stars. Extrapolated across the whole sky, that density implies on the order of tens of billions to over a hundred billion galaxies in the observable universe — later deep fields and modeling have pushed the census still higher.

Crucially, the galaxies did not all look like the grand spirals and smooth ellipticals near us. Many were small, irregular, blue, and clumpy — the disorderly building blocks of galaxy assembly seen when the universe was young. Because the light from the faintest objects had travelled for billions of years, the HDF is not just a map of space but a core sample through cosmic time: nearby, mature galaxies in the foreground; increasingly primitive, high-redshift systems fading into the deep background. The most distant confirmed galaxies in the field lie at redshifts beyond z ≈ 3–6, their light emitted when the cosmos was a small fraction of its present age.

Redshift, lookback time, and reading the field as history

The deep field works as a time machine because of two facts. First, light travels at a finite speed, so distant objects are seen as they were long ago — lookback time. Second, the expansion of the universe stretches the wavelength of that light on its way to us, an effect quantified by cosmological redshift, z. A galaxy at z = 1 is seen as it was roughly 8 billion years ago; z ≈ 6 corresponds to under a billion years after the Big Bang. As z rises, a galaxy's ultraviolet and blue light is shifted into the red and near-infrared.

• Because of that shift, the faintest, most distant HDF galaxies appear redder and are best detected in the F814W (near-IR) frame — a first hint of why later deep fields leaned ever more heavily on infrared instruments.
• The four-band photometry let astronomers estimate distances for thousands of galaxies at once using photometric redshifts — colors as a coarse proxy for spectra — without needing to take an individual spectrum of each faint smudge.
• The distribution of galaxies with redshift became a direct test of galaxy evolution and of Hubble's law, the relation between distance and recession velocity that gave the telescope its name.

The HDF did not just count galaxies; it let astronomers watch how the population changed with epoch — how star-formation rates, sizes, and shapes evolved over most of cosmic history.

Why it mattered, and the fields that followed

The Hubble Deep Field transformed a philosophical intuition — that the universe looks the same in every direction — into a hard, quantitative dataset, and it did something unusual for its era: the entire reduced dataset was released to the community immediately and publicly. Hundreds of research groups worldwide mined it, producing an avalanche of papers on galaxy counts, evolution, and the cosmic star-formation history. It became one of the most productive single observations in the history of astronomy.

Its success launched a lineage of ever-deeper campaigns:

• HDF-South (1998) pointed at a southern field near a distant quasar, confirming the northern result was no statistical fluke — the sky really is uniformly full of galaxies.
• The Hubble Ultra Deep Field (2003–04) used the far more sensitive Advanced Camera for Surveys and NICMOS in the constellation Fornax, gathering about a million seconds of exposure to reveal roughly 10,000 galaxies down to ~31st magnitude.
• The eXtreme Deep Field (2012) combined ten years of exposures on that same patch — nearly 2 million seconds — for the deepest optical image Hubble ever made.
• Beginning in 2022, the James Webb Space Telescope revisited deep-field territory in the infrared, pushing to redshifts z ≳ 13 and imaging galaxies from the first few hundred million years after the Big Bang.

Common misconceptions and look-alikes

"Those are stars." Almost none are. In the HDF the crisp, spike-bearing points are the rare foreground Milky Way stars; virtually every other object — even the faint smudges — is an entire galaxy. The diffraction spikes belong only to genuine point sources bright enough to reveal the telescope's optics.

"Hubble zoomed in with huge magnification." Depth here comes from integration time and light-gathering, not magnification. The field is tiny because Hubble's camera has a small field of view at high resolution, but the faintness is bought purely by staring — collecting photons over 100+ hours and rejecting noise by stacking.

"It's a single photograph." The final image is a composite of 342 dithered exposures in four filters, cleaned of cosmic rays and drizzled together. No single frame shows anything like the depth of the release image.

"The deep field is a rare, special direction." The opposite is true, and that is the whole point: the direction was chosen for being utterly ordinary. Point Hubble at almost any dark, high-latitude patch and, given enough time, you would find the same crowd of galaxies. The HDF's greatest lesson is that there is no empty sky — only sky we have not yet looked at long enough.

The deep-field campaigns compared: each pushed fainter and further into cosmic time.

Field	Year	Instruments	Depth / yield	Note
Hubble Deep Field (HDF-N)	1995	WFPC2 (UV–I)	~30 mag, ~3,000 galaxies	Ursa Major, near Big Dipper
HDF-South	1998	WFPC2 + STIS + NICMOS	Comparable depth	Confirmed HDF wasn't a fluke
Hubble Ultra Deep Field	2003–04	ACS + NICMOS	~31 mag, ~10,000 galaxies	Fornax; ~1 million sec
eXtreme Deep Field (XDF)	2012	ACS + WFC3/IR	~31.2 mag, ~5,500 galaxies	10-yr HUDF stack, ~2 Msec
JWST deep fields	2022–	NIRCam + MIRI	z ≳ 13 galaxies	Infrared; z~14 candidates