X-ray fluorescence, or XRF, investigates elemental composition by measuring characteristic X-rays emitted after a sample is excited. It is useful across materials, environmental and heritage studies, but a recognisable peak is not yet a complete quantitative composition.
The gold standard follows the entire inference: excitation, emitted radiation, detector response, spectral interpretation and calibration. This root guide develops the common principles; the linked archaeology guide handles the narrower problem of portable XRF on heritage materials.
eduKateSG · X-ray Fluorescence
Find your next learning step
Move from an elemental peak to a defensible result, keeping counting time, background and matrix effects visible.
Full chapter index · Practice and transfer · Education root · How X Works
Full chapter index
Understand the emission · Chapters 1–2
Work the counts · Chapters 3–4
Test the result · Chapters 5–7
Connect and practise · Chapters 8–10
Find specialist routes · Chapters 11–12
CHAPTER 1 OF 12 · Understand the emission
1. Follow excitation and relaxation separately
An incident X-ray can remove an inner-shell electron when sufficient energy is available. A higher-level electron may then fill the vacancy, with a characteristic X-ray emitted in a radiative relaxation. Other relaxation pathways also exist.
The incoming photon energy and the emitted line energy play different roles. The source must enable the interaction; the emission energy helps identify the transition and element. They should not be drawn as one unchanged photon passing through the sample.
Begin with three labelled stages: excite, emit, detect. This simple separation makes it easier to explain why an XRF measurement is not the same as a transmission image.
CHAPTER 2 OF 12 · Understand the emission
2. Energy identifies candidates, not a complete specimen
An energy-dispersive spectrum records counts across detector energy channels. Candidate emission lines help identify elements in the illuminated region. Peak assignment should consider calibration, resolution and other lines expected from the same element.
The specimen’s name is a further interpretation. A copper signal does not alone prove pure copper, a particular alloy recipe or a unique source. A measured area might include a coating, corrosion or several phases.
Ask two questions in order: which elemental emissions are supported by the spectrum, and what material claims do those observations justify? Do not jump straight from a coloured peak label to the history of the whole object.
Suppose a selected region contains 1,800 gross counts collected over 60 s. A separate representative background measurement gives 600 counts over 120 s. The gross rate is 30 counts/s and the background rate is 5 counts/s.
The estimated net rate is therefore 25 counts/s. Equivalently, scale the background to 60 s: 600 × 60/120 = 300 counts, giving 1,500 net counts in the sample acquisition.
Subtracting 600 directly from 1,800 would mix durations. This hypothetical calculation assumes the background is transferable and the acquisition times are appropriate corrected times; it does not solve overlapping peaks or detector nonlinearity.
Consider an invented calibration for a fixed matrix and acquisition setup: net rate S = 2.0c + 1.0, where S is counts/s and c is concentration in an arbitrary stated unit. For S = 25 counts/s, c = (25 − 1)/2.0 = 12 units.
The intercept is part of this fitted model; it is not automatically the same quantity as the separately measured background. Define the calibration response before applying both corrections, otherwise a contribution may be removed twice.
Now imagine the matrix-specific sensitivity were 1.5 rather than 2.0, with the same intercept. The result would be 16 units. That difference is a model sensitivity example: one count rate does not establish concentration independently of calibration.
CHAPTER 5 OF 12 · Test the result
5. Peak height and peak area answer different questions
A detector broadens emission lines into finite peaks. Two spectra can have similar integrated counts but different peak heights if their widths differ. Comparing only the highest channel can therefore misrepresent their relative signals.
Peak overlaps also matter. Larch’s XRF documentation describes overlapping line families and detector artefacts such as dead time and pileup. A fit must distinguish those contributions rather than treating every bump as a new element.
For a learner, sketch two peaks with equal areas and different widths. Explain why the taller one need not represent more emitted photons. Then add a sloping background and identify the additional quantity that must be estimated.
The matrix is the rest of the sample around the element of interest. Absorption and enhancement can alter the radiation reaching the detector. Physical differences such as particle size and surface condition can also affect comparability.
This makes ‘same instrument, same element’ an incomplete comparison. The calibration needs to address the sample type and geometry for the intended claim. An internal standard or a physical model can help in suitable cases, but neither is a universal correction for every mismatch.
A useful explanation says what was held constant, what was corrected and what remains different. If an unknown lies outside the validated sample conditions, the output should retain that limitation rather than borrow confidence from a familiar instrument name.
| Evidence stage | What it supports | What still needs checking |
|---|---|---|
| Line assignment | Candidate elemental presence | Calibration, resolution and overlaps |
| Net signal | A corrected spectral response | Background model and detector behaviour |
| Quantitative result | Concentration under a method | Matrix, standards and validation |
CHAPTER 7 OF 12 · Test the result
7. Validate with information not used to fit the result
A calibration can reproduce its own standards and still fail on another material. Use appropriate independent reference or check samples to test the intended range and matrix. Repeat measurements alone mostly assess consistency under the repeated conditions.
Suppose an independent check material has a reference value of 20 units and the method returns 22. The relative difference is (22 − 20)/20 × 100 = +10%. That is a useful diagnostic, not an automatic universal pass or fail.
State the check’s uncertainty and the acceptance criterion appropriate to the purpose. Do not invent a tolerance after seeing the result. If the check is unsuitable for the unknown, say so before interpreting agreement as validation.
An XRF map associates spectra or fitted signals with scan positions. It can reveal where elemental signals differ across the examined area. A bright pixel still reflects excitation, attenuation and acquisition conditions as well as composition.
A teaching audit should retain the scale bar, pixel size, acquisition conditions and quantity mapped. A map of raw counts is not automatically a map of concentration, and a surface-area map does not establish the composition of an entire object.
Compare like with like before interpreting a boundary as a material interface. If the question concerns the bulk specimen, identify what additional sampling would make that claim representative.
CHAPTER 9 OF 12 · Connect and practise
9. Choose fluorescence, absorption or diffraction by the question
XRF usually serves an elemental emission question. X-ray absorption spectroscopy follows how absorption changes with incident energy and can investigate an element’s local environment. Diffraction investigates structural ordering through scattered-wave interference.
These methods can complement one another. Agreement should be described in terms of the separate evidence each provides, not as three versions of the same measurement.
For a specimen with an unknown coating, begin by defining whether the priority is elemental distribution, chemical environment or crystalline phases. The question should choose the method and sampling plan; a convenient instrument should not silently redefine the question.
First, calculate a net rate from 2,400 gross counts over 80 s and 900 representative background counts over 150 s. Second, apply the teaching calibration S = 3c + 2 to that net rate.
Checks: gross rate = 30 counts/s, background rate = 6 counts/s and net rate = 24 counts/s. Then c = (24 − 2)/3 ≈ 7.33 concentration units, within the explicitly assumed model.
Now keep the rates but change the sample from the calibrated matrix to an unvalidated coating. The arithmetic remains 7.33, but the quantitative interpretation is not established. Explain which check sample or alternate method would be needed before treating it as a reliable coating concentration.
Develop the spectral reasoning
- Characteristic lines and elemental evidence — follow the XRF mechanism and its evidence limits.
- X-ray Spectroscopy — compare emission with absorption and photoelectron methods.
Transfer into a specific application
- Portable XRF, calibration standards and matrix effects — apply the reasoning to archaeological surfaces.
- Synchrotron light and beamlines — understand a source and experimental environment used for element-specific studies.
Read the heritage guide for sample-surface and provenance questions. Stay with the spectroscopy root when the confusion is which interaction produced the recorded signal.
- U.S. EPA CLU-IN: X-ray fluorescence — mechanism, environmental applications, calibration and interferences.
- Larch: X-ray fluorescence analysis — spectra, peak overlap, attenuation and detector artefacts.
- USGS: internal standards in X-ray spectroscopy — a research example of matrix correction and its limits.
The numerical calibrations and specimens here are invented for teaching. They are not a validated analytical method or a safety determination. X-ray equipment belongs in appropriately controlled settings with trained personnel; the learning exercises require no radiation source.
The final checkpoint is to explain how a measured count became a material claim, and where an unsupported assumption would enter. That explanation is the bridge back to the Education root’s principle of transferable learning.